RU2777926C2 - Gan-cnn for prediction of mhc-peptide binding - Google Patents

Abstract

FIELD: biotechnology.

SUBSTANCE: methods for training a generative-adversarial network (hereinafter – GAN) in combination with a convolutional neural network (hereinafter – CNN) are disclosed. GAN and CNN can be trained using biological data, such as protein interaction data. CNN can be used to identify new data as positive or negative. Methods for the synthesis of polypeptide associated with new data on interaction of proteins identified as positive are disclosed.

EFFECT: GAN-CNN prediction of MHC-peptide binding.

25 cl, 17 dwg, 2 tbl, 6 ex

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present application claims priority under U.S. Provisional Application No. 62/631,710, filed Feb. 17, 2018, and hereby incorporated herein by reference in its entirety.

BACKGROUND OF THE INVENTION

[0002] One of the biggest challenges facing machine learning is the lack of large annotated datasets. Data annotations are not only costly and time consuming, but also highly dependent on the availability of expert observers. A limited amount of training data can reduce the effectiveness of supervised machine learning algorithms, which often require very large amounts of training data to avoid overfitting. Until recently, a lot of effort has gone into extracting as much information as possible from the available data. One area in particular lacking in large annotated datasets is the analysis of biological data, such as protein interaction data. The ability to predict how proteins might interact is invaluable for identifying new therapeutics.

[0003] Advances in immunotherapy are advancing rapidly and provide new drugs that modulate the patient's immune system to help fight disease, including cancer, autoimmune disorders, and infections. For example, immune checkpoint inhibitor molecules, such as PD-1 and PD-1 ligands, have been identified and are being used to develop drugs that inhibit or stimulate signaling through PD-1 and thereby modulate the patient's immune system. These new drugs have proven to be very effective in some cases, but not in all. One reason is that about 80 percent of cancer patients' tumors do not have enough cancer antigens to attract T cells.

[0004] Targeting tumor-specific mutations in an individual is promising because such specific mutations produce tumor-specific peptides called neoantigens that are new to the immune system and are not found in normal tissues. Compared to tumor-associated autoantigens, neoantigens elicit T-cell responses that are independent of the host's central thymus-mediated tolerance and also elicit less autoimmune-mediated toxicity against benign cells (Nature Biotechnology 35, 97 (2017).

[0005] A key question for the discovery of the neoepitope is which mutated proteins are processed by the proteasome into peptides of 8-11 residues, translocated to the endoplasmic reticulum via an antigen processing-associated transporter (TAP), and loaded onto the newly synthesized MHC class I (MHC-I) for CD8+ T cell recognition (Nature Biotechnology 35, 97 (2017)).

[0006] Computational methods for predicting the interaction of a peptide with MHC-I are known in the art. Although some computational methods focus on predicting what happens during antigen processing (eg NetChop) and peptide transport (eg NetCTL), most efforts are focused on modeling which peptides bind to the MHC-I molecule. Neural network based methods such as NetMHC are used to predict antigen sequences that form epitopes corresponding to the cavity of a patient's MHC-I molecules. Other filters can be used to deprioritize hypothetical proteins and determine whether the mutated amino acid is oriented so that it likely faces outward from the MHC (towards the T-cell receptor) or reduces the epitope's affinity for the MHC-I molecule itself (Nature Biotechnology 35 , 97 (2017).

[0007] These predictions may be incorrect for many reasons. Sequencing already introduces amplification and technical errors into the reads used as starting material for peptides. When modeling epitope processing and presentation, it is also necessary to take into account the fact that humans have ∼ 5000 alleles encoding MHC-I molecules, while up to six of them are expressed in an individual patient, all with different epitope affinities. Methods such as NetMHC typically require 50-100 experimentally determined peptide binding measurements for a particular allele to build a model with sufficient accuracy. But because many MHC alleles lack such data, "pan-specific" methods - capable of predicting binding based on whether MHC alleles with similar contact environments have similar binding specificities - are increasingly coming to the fore.

[0008] Thus, there is a need for improved systems and methods for generating datasets for use in machine learning applications, in particular biological datasets. Methods for predicting peptide binding may benefit from such improved systems and methods. Thus, it is an object of the present invention to provide computer-based systems and methods that have improved dataset generation capabilities for training machine learning applications to make predictions, including predicting peptide binding to MHC-I.

SHORT DESCRIPTION

[0009] It is to be understood that both the following general description and the following detailed description are illustrative and explanatory only, and are not limiting.

[0010] Methods and systems for training a generative adversarial network (GAN) are provided, providing for the generation by the GAN generator of increasingly accurate positive simulated data until the GAN discriminator classifies positive simulated data as positive, representing positive simulated data, positive real data and negative real data into a convolutional neural network (CNN) until the CNN classifies each type of data as positive or negative, representing positive real data and negative real data in the CNN to generate prediction scores, determining based on the prediction scores whether whether the GAN is trained or not, and derivation of the GAN and CNN. This method can be repeated until the GAN is properly trained. Positive simulated data, positive real data, and negative real data include biological data. The biological data may include protein-protein interaction data. The biological data may include polypeptide-MHC-I interaction data. Positive simulated data may include positive simulated polypeptide-MHC-I interaction data, positive real data include positive real polypeptide-MHC-I interaction data, and negative real data include negative real polypeptide-MHC-I interaction data.

[0011] Additional advantages will be set forth in part in the following description, or may be learned from practice. The advantages will be realized and achieved by the features and combinations specifically set forth in the appended claims.

BRIEF DESCRIPTION OF GRAPHICS

[0012] The accompanying drawings, which are included in and form part of this specification, illustrate embodiments and, together with the description, serve to explain the principles of the methods and systems.

The figure 1 presents a block diagram illustrating an example of the method.

Figure 2 is an exemplary diagram showing part of the peptide binding prediction process, including generation and training of GAN models.

Figure 3 is an exemplary diagram showing part of the peptide binding prediction process, including generating data using trained GAN models and training CNN models.

Figure 4 is an exemplary diagram showing part of the peptide binding prediction process, including completing the training of the CNN models and generating peptide binding predictions using the trained CNN models.

Figure 5A is an exemplary data flow diagram of a typical GAN.

Figure 5B is an exemplary data flow diagram of a GAN generator.

Figure 6 is an exemplary block diagram of a portion of the processing steps provided by the generator used in the GAN.

Figure 7 is an exemplary block diagram of a portion of the processing steps provided by the generator used in the GAN.

Figure 8 is an exemplary block diagram of a portion of the processing steps provided by a discriminator used in a GAN.

Figure 9 is an exemplary block diagram of a portion of the processing steps provided by a discriminator used in a GAN.

The figure 10 presents a flowchart illustrating an example of the method.

Figure 11 is an exemplary block diagram of a computer system that can implement the processes and structures involved in predicting peptide binding.

Figure 12 is a table showing the results of these predictive models for predicting protein binding to the MHC-I protein complex for the indicated HLA alleles.

Figure 13A is a table showing the data used to compare prediction models.

Figure 13B is a histogram comparing the AUC of the present implementation of the CNN architecture and the same one from Wang's article.

Figure 13C is a bar chart comparing the present implementation with existing systems.

Figure 14 is a table showing the bias obtained by selecting a biased test set.

Figure 15 is a line plot of SRCC versus test size, showing that the smaller the test size, the better the SRRC.

Figure 16A is a table showing the data used to compare the Adam and RMSprop neural networks.

Figure 16B is a histogram comparing the AUC of neural networks trained by the Adam optimizer and RMSprop.

Figure 16C is a histogram of the SRCC comparison of neural networks trained by the Adam optimizer and RMSprop.

Figure 17 is a table showing that the combination of fake and real data provides better prediction than fake data alone.

DETAILED DESCRIPTION OF THE INVENTION

[0013] Before the present methods and systems are disclosed and described, it should be understood that the methods and systems are not limited to specific methods, specific components, or specific implementations. It should also be understood that the terminology used herein is for the purpose of describing certain embodiments only and is not intended to be limiting.

[0014] As used in the description and the appended claims, the singular form includes references to the plural, unless the context clearly requires otherwise. Ranges may be expressed herein as from "about" one specific value and/or to "about" another specific value. Where such a range is expressed, then another embodiment includes from one specific value and/or to another specific value. Similarly, when values are expressed as approximations using the preceding "approximately", then it will be understood that a particular value constitutes another embodiment. It will also be understood that the endpoints of each of the ranges are significant both with respect to the other endpoint and independently of the other endpoint.

[0015] "Optional" or "optional" means that the event or circumstance described below may or may not occur, and that the description includes cases in which the specified event or circumstance occurs and cases in which it does not occur.

[0016] Throughout the description and claims of this document, the word "comprise" and variants of this word, such as "comprising" and "comprises", means "including without limitation", and is not intended to exclude, for example, other components, whole numbers or stages. The expression "exemplary" means "example" and is not intended to indicate a preferred or ideal embodiment. The expression "such as" is not used in a restrictive sense, but for explanatory purposes.

[0017] It is understood that the methods and systems are not limited to the particular methodology, protocols, and reagents described, as they may vary. It should also be understood that the terminology used herein is only intended to describe specific embodiments and is not intended to limit the scope of the present methods and system, which will be limited only by the appended claims.

[0018] Unless otherwise defined herein, all technical and scientific terms used herein have the same meanings as generally understood by a person skilled in the art to which the methods and systems belong. While any methods and materials similar to or equivalent to those described herein may be used in the practice or testing of the present method or compositions, only the most preferred methods, devices and materials are described herein. The publications cited in this document and the materials to which they are cited are hereby expressly incorporated by reference. Nothing herein should be construed as an admission that the present methods and systems are not entitled to precede such disclosure by virtue of prior invention. No admission is made that any references constitute prior art. Discussions of references indicate what their authors claim, and applicants reserve the right to challenge the accuracy and timeliness of cited documents. It will be clearly understood that although a number of publications are referred to in this document, such reference is not an admission that any of these documents form part of the general knowledge in this field.

[0019] This document discloses components that can be used to perform these methods and systems. These and other components are disclosed herein, and it is known that in the disclosure of combinations, subsets, interactions, groups, etc. of such components, although specific reference to each of the various individual and cumulative combinations and permutations of such components may not be clearly described herein, each of which is specifically contemplated and described herein in relation to all methods and systems. This applies to all embodiments of this application, including steps in these methods. Thus, if there are a number of additional steps that can be performed, it should be understood that each of these additional steps can be performed with any particular embodiment or combination of embodiments of the disclosed methods.

[0020] The present methods and systems may be more readily understood by referring to the following detailed description of the preferred embodiments and the examples included therein, as well as to the figures and their previous and following descriptions.

[0021] Methods and systems may take the form of an all-hardware implementation, an all-software implementation, or an implementation that combines software and hardware implementations. In addition, the methods and systems may take the form of a computer program product on a computer-readable storage medium having computer-readable program instructions (eg, computer software) embodied on the storage medium. More specifically, the present methods and systems may take the form of computer software implemented over the Internet. Any suitable computer-readable storage medium may be used, including hard disks, CD-ROMs, optical storage devices, or magnetic storage devices.

[0022] Embodiments of methods and systems are described below with reference to block diagrams and flowcharts illustrating methods, systems, devices, and computer program products. It will be appreciated that each block of block diagram and block diagram illustrations and combinations of blocks in the block diagram and block diagram illustrations, respectively, can be implemented by computer program instructions. These computer program instructions can be downloaded to a general purpose computer, a special purpose computer, or other programmable data processing device, which is a machine, due to which the instructions executed by the computer or other programmable data processing device provide the implementation of the functions indicated in the block or blocks. -scheme.

[0023] Such computer program instructions may also be stored in a computer-readable memory device that can control a computer or other programmable data processing device to perform its functions properly, whereby the instructions stored in the computer-readable memory device provide a finished product, including computer-readable commands intended to implement the function indicated in the block or blocks of the block diagram. Computer program instructions may also be downloaded to a computer or other programmable processing device to cause a number of process steps to be performed on the computer or other programmable device to form a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementation of the functions indicated in the block or blocks of the block diagram.

[0024] Accordingly, block diagrams and block diagrams provide combinations of means to perform said functions, combinations of steps to perform said functions, and means represented by program instructions to perform said functions. It should also be understood that each block diagram and block diagram illustration and combination of block diagram and block diagram illustrations can be implemented with special purpose computer hardware systems that perform the indicated functions or steps or combinations. specialized hardware and computer programs.

I. Definitions

[0025] The abbreviation "SRCC" refers to Spearman Rank Correlation Coefficient (SRCC) calculations.

[0026] The term "ROC curve" refers to a receiver performance curve.

[0027] The abbreviation "CNN" refers to a convolutional neural network.

[0028] The abbreviation "GAN" refers to a generative adversarial network.

[0029] The term "HLA" refers to human leukocyte antigen. The HLA system or complex is a gene complex encoding major histocompatibility complex (MHC) proteins in humans. The major HLA class I genes are HLA-A, HLA-B, and HLA-C, while HLA-E, HLA-F, and HLA-G are minor genes.

[0030] The term "MHC I" or "major histocompatibility complex I" refers to a set of cell surface proteins consisting of an α-chain having three domains - α1, α2 and α3. The α3 domain is a transmembrane domain, while the α1 and α2 domains are responsible for the formation of the peptide-binding cavity.

[0031] "Polypeptide-MHC I interaction" refers to the binding of a polypeptide in the peptide-binding cavity of the MHC I.

[0032] As used herein, the term "biological data" means any data obtained from the measurement of the biological conditions of humans, animals, or other biological organisms, including microorganisms, viruses, plants, and other living organisms. Measurements may be made using any test, analysis, or observation known to physicians, scientists, diagnosticians, and the like. Biological data may include, without limitation, DNA sequences, RNA sequences, protein sequences, protein interactions, clinical tests and observations, physical and chemical measurements, genomic determinations, proteomic determinations, drug levels, hormonal and immunological tests, neurochemical or neurophysical measurements, mineral and vitamin determinations, genetic and family history, and other determinations that may provide insight into the condition of the person or persons being tested. In this document, the term "data" is used interchangeably with the term "biological data".

II. Peptide binding prediction systems

[0033] In one embodiment, the present invention provides a system for predicting peptide binding to MHC-I, which has a generative adversarial network (GAN)-convolutional neural network (CNN) structure, also referred to as a deep convolutional generative adversarial network. The GAN contains a CNN discriminator and a CNN generator and can be trained from existing peptide-MHC-I binding data. The disclosed GAN-CNN systems have several advantages over existing systems for predicting peptide-MHC-I binding, including, but not limited to, the ability to learn through an unlimited number of alleles and better prediction performance. The present methods and systems, although described herein in relation to predicting peptide binding to MHC-I, are not limited to the ways in which these methods and systems can be used. The prediction of peptide binding to MHC-I is presented as an example of the application of the improved GAN-CNN system described herein. The improved GAN-CNN system is applicable to a wide range of biological data to obtain various predictions.

A. Exemplary Neural Network Systems and Methods

[0034] Figure 1 is a flowchart 100 illustrating an example of a method. From step 110 , the GAN generator (see 504 in FIG. 5A ) can generate increasingly accurate positive simulated data. Positive simulated data may include biological data, such as protein interaction data (eg, binding affinity). Binding affinity is one example of measuring the strength of a binding interaction between a biomolecule (eg, protein, DNA, drug, etc.) and a biomolecule (eg, protein, DNA, drug, etc.). Binding affinity can be expressed numerically as the value of the half-maximal inhibitory concentration (IC ₅₀ ). A lower number indicates a higher affinity. Peptides with IC50 values <50 nM are considered high affinity, those with <500 nM intermediate affinity, and <5000 nM low affinity. IC ₅₀ can be converted to a binding category as binding (1) or no binding (-1).

[0035] Positive simulated data may include positive simulated polypeptide-MHC-I interaction data. The creation of positive simulated polypeptide-MHC-I interaction data can be based, at least in part, on actual polypeptide-MHC-I interaction data. The protein interaction data may include a binding affinity score (eg, IC ₅₀ , binding category) representing the likelihood of two proteins binding. Protein interaction data, e.g., polypeptide-MHC-I interaction data, can be obtained, for example, from any number of databases such as PepBDB, PepBind, Protein Data Bank, Biomolecular Interaction Network Database (BIND), Cellzome (Heidelberg, Germany). ), Interacting Protein Database (DIP), Dana Farber Cancer Institute (Boston, Massachusetts, USA), Human Protein Reference Database (HPRD), Hybrigenics (Paris, France), European Bioinformatics Institute (EMBL-EBI, Hinxton, UK) IntAct, Molecular Interaction Database (MINT, Rome, Italy), Protein-Protein Interaction Database (PPID, Edinburgh, UK) and Gene/Protein Interaction Retrieval Search Tool (STRING, EMBL, Heidelberg, Germany), etc. . The protein interaction data may be stored in a data structure containing one or more of a particular polypeptide sequence, as well as an indication regarding the interaction of the polypeptides (eg, the interaction between the polypeptide sequence and MHC-I). In one embodiment, the data structure may follow the HUPO PSI Molecular Interaction (PSI MI) format, which may contain one or more entries, where the entry describes one or more protein interactions. The data structure can point to the source of the entry, such as a data provider. The release number and release date assigned by the data provider may be specified. The availability list may contain statements about the availability of data. The list of experiments may include descriptions of the experiments, including at least one set of experimental parameters, usually associated with a particular publication. In large-scale experiments, typically only one parameter is varied in a series of experiments, often the primer (protein of interest). The PSI MI format can specify both constant parameters (eg, experimental procedure) and variable parameters (eg, primer). The list of partners may indicate the set of partners (eg proteins, small molecules, etc.) involved in the interaction. The partner protein element may indicate the "normal" form of the protein, commonly found in databases such as Swiss-Prot and TrEMBL, which may include data such as name, cross-references, organism, and amino acid sequence. The list of partners may indicate one or more interaction elements. Each interaction can point to an availability description (data availability description) and a description of the experimental conditions under which it was determined. An interaction can also indicate a confidence attribute. Various measures of confidence in interactions have been developed, such as the method for verifying paralogs and the biological evaluation of a protein interaction map (PIM). Each interaction may specify a list of participants, containing two or more protein elements-participants (ie, proteins involved in the interaction). Each participating protein element may include a description of the molecule in its native form and/or the specific form of the molecule in which it participated in the interaction. The list of characteristics may indicate features of the protein sequence, such as binding domains or post-translational modifications related to the interaction. A role can be specified that describes the specific role of the protein in the experiment, such as whether the protein was a primer or a prey. Some or all of the previous elements may be stored in a data structure. An example of a data structure would be an XML file, like this:

<entry>

<interactorList>

<Interactor id="Succinate>

<name>

<shortLabel>Succinate</shortLabel>

<fullName>Succinate</fullName>

</name>

</Interactor>

</interactorList>

<interactionList>

<interaction>

<name>

<shortLabel> Succinate dehydrogenas catalysis </shortLabel> <fullName>Interaction between </fullName>

</name>

<ParticipantList> <Participant>

<proteinInteractorRef ref="Succinate"/› <biologicalrole>neutral</role> </proteinParticipant> <proteinParticipant> <proteinInteractorRef ref="Fumarate"/> <role>neutral</role> </proteinParticipant> <proteinParticipant> <proteinInteractorRef ref="Succdeh"/> <role>neutral</role> </proteinParticipant> </participantList> </interaction>

</interactionList>

[0036] The GAN may include, for example, deep convolutional GAN (DCGAN). In FIG. 5A shows an example of a basic GAN structure. GAN is essentially a way to train a neural network. GANs typically contain two independent neural networks, a discriminator 502 and a generator 504 , that operate independently and can act as opposing parties. The discriminator 502 may be a neural network to be trained using the training data generated by the generator 504 . The discriminator 502 may include a classifier 506 that can be trained to perform the task of distinguishing between data instances. Generator 504 may generate random data instances similar to real instances, but which may be generated including characteristics, or may be modified to include characteristics that make them fake or fake instances. Neural networks comprising a discriminator 502 and a generator 504 can generally be implemented as multilayer networks consisting of multiple processing layers such as density processing, batch normalization processing, activation processing, input reshaping processing, Gaussian elimination processing. , Gaussian noise processing, 2D convolution, and 2D upsampling. This is shown in more detail in FIG. 6 - fig. 9 below.

[0037] For example, classifier 506 may be designed to identify data instances indicative of various characteristics. The generator 504 may include an adversarial function 508 that may generate data designed to mislead the discriminator 502 using data instances that are nearly correct, but not quite. For example, this can be done by randomly selecting a valid instance from the training set 510 (latent space) and synthesizing the data instance (data space) by randomly changing its characteristics, for example by adding random noise 512 . The generative network G can be considered as a transfer of some latent space to the data space. Formally, this can be expressed as G: G (z) → R ^|x| , where z ∈ R ^|x| is an instance from the hidden space, x ∈ R ^|x| is an instance from the data space, and | | denotes the number of measurements.

[0038] The discriminator network, D, can be thought of as a transfer from data space to the probability that the data (eg, peptide) is from a real data set and not from a generated (fake or fake) data set. Formally, this can be expressed as: D : D(x) → (0; 1). During training, discriminator 502 may be presented by randomizer 514 with a random combination of valid data instances 516 from real training data along with fake or fake (eg, simulated) data instances generated by generator 504 . For each instance of data, discriminator 502 may attempt to identify valid and fake or spurious inputs, returning a result of 518 . For example, for a fixed generator G, a discriminator D can be trained to classify data (eg, peptides) as coming from training data (real, close to 1) or from a fixed generator (simulated, close to 0). For each instance of data, discriminator 502 may additionally attempt to identify positive or negative inputs (regardless of whether the inputs are simulated or real), producing a result of 518 .

[0039] Based on the series of results 518 , both discriminator 502 and generator 504 may attempt to fine-tune their parameters to improve their performance. For example, if discriminator 502 makes a correct prediction, generator 504 may update its parameters to produce better modeled instances to mislead discriminator 502 . If the discriminator 502 makes an incorrect prediction, the discriminator 502 can learn from its error to avoid similar errors. Thus, updating the discriminator 502 and generator 504 may include a feedback process. This feedback process can be continuous or incremental. Generator 504 and discriminator 502 may operate iteratively to optimize data generation and data classification. In the incremental feedback process, the state of the generator 504 is latched and the discriminator 502 is trained until an equilibrium is established and the training of the discriminator 502 is optimized. For example, given a fixed state of generator 504 , discriminator 502 can be trained to be optimized with respect to the state of generator 504 . This optimized state of the discriminator 502 can then be captured and the generator 504 can be trained to reduce the accuracy of the discriminator to some predetermined threshold. Then the state of the generator 504 can be captured and the discriminator 502 can be trained, and so on.

[0040] In a continuous feedback process, the discriminator cannot be trained until its state is optimized, but can only be trained for one or a small number of iterations, and the generator can be updated at the same time as the discriminator.

[0041] If the distribution of the generated simulated dataset can perfectly match the distribution of the real dataset, then the discriminator will be as confused as possible and will not be able to distinguish between real and fake instances (prediction 0.5 for all inputs).

[0042] Returning to FIG. 1 to stage 110 ; generating increasingly accurate positive simulated polypeptide-MHC-I interaction data may be performed (eg, by generator 504 ) until the GAN discriminator 502 classifies positive simulated polypeptide-MHC-I interaction data as positive. In another aspect, generation of increasingly accurate positive simulated polypeptide-MHC-I interaction data may be performed (eg, by generator 504 ) until the GAN discriminator 502 classifies positive simulated polypeptide-MHC-I interaction data as positive. For example, generator 504 can generate progressively more accurate positive simulated polypeptide-MHC-I interaction data by generating a first simulated dataset containing positive simulated polypeptide-MHC-I interactions for the MHC allele. The first simulated dataset may be generated according to one or more GAN parameters. GAN parameters may include, for example, one or more of allele type (e.g., HLA-A, HLA-B, HLA-C, or subtypes thereof), allele length (e.g., about 8 to 12 amino acids, about 9 to 11 amino acids ), generation category, model complexity, learning rate, batch size, or other parameter.

[0043] FIG. 5B is an exemplary data flow diagram of a GAN generator configured to generate positive simulated polypeptide-MHC-I interaction data for an MHC allele. As shown in FIG. 5B , a Gaussian noise vector may be input to a generator that outputs a distribution matrix. Gaussian sampled input noises provide variability that mimics different binding patterns. The output distribution matrix is the probability distribution of the selection of each amino acid for each position in the peptide sequence. The distribution matrix can be normalized to get rid of variants that are less likely to provide binding signals, and a particular peptide sequence can be taken from the normalized distribution matrix.

[0044] The first simulated data set can then be combined with positive real polypeptide interaction data and/or negative real polypeptide interaction data (or a combination thereof) for the MHC allele to create a GAN training set. The discriminator 502 can then determine (eg, according to a decision boundary) whether the polypeptide-MHC-I interaction for the MHC allele in the GAN training dataset is positive or negative and/or simulated or real. Based on the accuracy of the determination made by the discriminator 502 (eg, whether the discriminator 502 correctly identified the polypeptide-MHC-I interaction as positive or negative and/or simulated or real), one or more GAN parameters or a decision boundary may be adjusted. For example, one or more GAN decision boundary parameters can be adjusted to optimize the discriminator 502 to increase the ability to assign a high probability to positive real polypeptide-MHC-I interaction data, a low probability to positive simulated polypeptide-MHC-I interaction data, and/or a low probability to negative real data of polypeptide-MHC-I interaction. One or more of the decision boundary GAN parameters can be adjusted to optimize generator 504 to increase the likelihood that positive simulated polypeptide-MHC-I interaction data will score high.

[0045] The process of generating a first simulated data set, combining the first data set with positive real polypeptide interaction data and/or negative real polypeptide interaction data to generate a GAN training data set, determining with a discriminator and adjusting the GAN parameters and/or decision boundary can be repeat until the first stop criterion is met. For example, you can determine if the first stopping criterion is met by evaluating the gradient descent expression for generator 504 . As another example, one can determine if the first stopping criterion is met by evaluating the root mean square error (MSE) function:

[0046] As another example, it can be determined whether the first stopping criterion is met by evaluating whether the gradient is large enough to continue meaningful learning. Since the generator 504 is updated with the backpropagation algorithm, each layer of the generator will have one or more gradients, for example, given that the plot has 2 layers and each layer has 3 nodes, plot output 1 is 1-dimensional (scalar) and the data are 2-dimensional. In this graph, the ^1st layer has 2*3=6 edges (w111, w112, w121, w122, w131, w132) connected to the data and w111*data1+w112*data2=net11 and a sigmoid activation function can be used to deriving o11=sigmoid(net11), similarly, you can get o12, o13, which form the output data of the ^1st layer; The ^2nd layer has 3*3=9 edges (w211, w212, w213, w221, w222, w223, w231, w232, w233) connecting to the output of the ^1st layer, and the output of the ^2nd layer is o21 , o22, o23, and connects the final output to 3 edges which are w311, w312, w313.

[0047] Each w in this graph has a gradient (a command on how to update w is essentially a number to be added), the number can be calculated using an algorithm called backpropagation, following the idea of changing the parameter in the direction where loss (MSE) decreases, which is:

Where E is the MSE error, w _ij is the i -th parameter on the j -th layer. O _j is the output on the j -th layer, net _j - before activation, the result of multiplication on the j -th layer. And if the value of de / dw _ij (gradient) for w _ij is not large enough, as a result, training does not change for w _ij generator 504 , and training should be stopped.

[0048] Then, after the GAN discriminator 502 classifies the positive simulated data (e.g., positive simulated polypeptide-MHC-I interaction data) as positive and/or real, at step 120 positive simulated data, positive real data, and/or negative real data. data (or a combination of them) can be represented in the CNN as long as the CNN does not classify each type of data as positive or negative. Positive simulated data, positive real data, and/or negative real data may constitute biological data. Positive simulated data may include positive simulated polypeptide-MHC-I interaction data. Positive real data may include positive real data for polypeptide-MHC-I interaction. Negative real data may include negative real data for polypeptide-MHC-I interaction. Classified data may include data regarding the interaction of the polypeptide-MHC-I. Each of positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data can be associated with a selected allele. For example, the selected allele may be selected from the group consisting of A0201, A202, A203, B2703, B2705, and combinations thereof.

[0049] Representing positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data in a CNN may include generating, for example, using a generator 504 according to a set of parameters GAN, a second set of simulated data containing positive simulated polypeptide-MHC-I interactions for the MHC allele. The second simulated data set can then be combined with positive real polypeptide interaction data and/or negative real polypeptide interaction data (or a combination thereof) for the MHC allele to create a CNN training data set.

[0050] The CNN training dataset may then be presented to the CNN for training the CNN. The CNN may then classify, according to one or more CNN parameters, the polypeptide-MHC-I interaction as positive or negative. This may include performing a convolution procedure by the CNN, performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure, and/or performing a classification procedure (eg, a fully connected layer).

[0051] Based on the classification accuracy of the CNN, one or more parameters of the CNN may be adjusted. The process of generating a second simulated dataset, generating a CNN training dataset, classifying the polypeptide-MHC-I interaction, and adjusting one or more CNN parameters may be repeated until the second stopping criterion is met. For example, you can determine if the second stopping criterion is met by evaluating the mean squared error (MSE) function:

[0052] Next, at step 130 , positive real data and/or negative real data may be presented to the CNN to generate prediction scores. Positive real data and/or negative real data may contain biological data, such as protein interaction data, including, for example, binding affinity data. Positive real data may include positive real data for polypeptide-MHC-I interaction. Negative real data may include negative real data for polypeptide-MHC-I interaction. The prediction scores may be binding affinity scores. Prediction scores may include the likelihood that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data. This may include presenting the actual data set to the CNN and classifying by the CNN according to the CNN parameters of the polypeptide-MHC-I interaction for the MHC allele as positive or negative.

[0053] At step 140 , it can be determined whether the GAN has been trained based on the prediction scores. This may include determining if the GAN is trained by determining the accuracy of the CNN based on the prediction scores. For example, a GAN can be defined as trained if the third stopping criterion is met. Determining whether the third stopping criterion is met may include determining whether the area under the curve (AUC) function is met. Determining whether the GAN has been trained may include comparing one or more prediction scores to a threshold. If the GAN is trained as determined in step 140 , then the GAN may be further inferred in step 150 . If the GAN is not determined to be trained, the GAN can be returned to step 110 .

[0054] After training the CNN and the GAN, the data set (eg, the unclassified data set) can be represented in the CNN. The dataset may contain unclassified biological data, such as unclassified protein interaction data. Biological data may include many candidate polypeptide-MHC-I interactions. The CNN can generate a predicted binding affinity and/or classify each of the candidate polypeptide-MHC-I interactions as positive or negative. A polypeptide can then be synthesized using candidate polypeptide-MHC-I interactions classified as positive. For example, the polypeptide may contain a tumor-specific antigen. As another example, the polypeptide may contain an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele.

[0055] A more detailed exemplary flow diagram of a prediction process 200 using a generative adversarial network (GAN) is shown in FIG . 2 - fig. 4 . Steps 202-214 generally correspond to step 110 shown in FIG. 1 . Process 200 may begin at 202 in which GAN training is configured, such as by setting a set of parameters 204-214 to control GAN training 216 . Examples of parameters that can be set may include allele type 204 , allele length 206 , generation category 208 , model complexity 210 , learning rate 212 , and batch size 214 . The allele type parameters 204 may provide the ability to specify one or more allele types to be included in the GAN processing. Examples of these types of alleles are shown in Fig. 12 . For example, these alleles may include A0201, A0202, A0203, B2703, B2705, etc. shown in FIG. 12 . The allele length parameters 206 may provide the ability to specify the lengths of peptides that can bind to each specified allele type 204 . Examples of such allele lengths are shown in FIG. 13 . For example, for A0201, the specified length is displayed as 9 or 10, for A0202, the specified length is displayed as 9, for A0203, the specified length is displayed as 9 or 10, for B2705 , the specified length is displayed as 9, and so on. categories of data to be generated during GAN training 216 . For example, binding/non-binding categories can be specified. A set of parameters corresponding to model complexity 210 may provide the ability to define aspects of model complexity to be used during GAN training 216 . Examples of such aspects may include the number of layers, the number of nodes per layer, the window size for each convolutional layer, etc. The learning rate parameters 212 may provide the ability to specify one or more rates at which the learning processing performed in GAN training 216 should converge. . Examples of such learning rate parameters may include 0.0015, 0.015, 0.01, which are dimensionless values defining relative learning rates. The packet size parameters 214 may provide the ability to specify the sizes of training data packets 218 to be processed during GAN training 216 . Examples of such packet sizes may include packets containing 64 or 128 data instances. GAN training setup processing 202 may collect training parameters 204-214 , process them to be compatible with GAN training 216 , and input the processed parameters into GAN training 216 or store the processed parameters in appropriate files or locations for use in GAN training 216 .

[0056] At 216 , GAN training may begin. Steps 216-228 also generally correspond to step 110 shown in FIG. 1 . Training 216 GAN can receive data 218 for training, for example, in batches, as indicated by the parameters 214 packet size. The training data 218 may include data representing peptides with different binding affinity designations (bind or not) for MHC-I protein complexes encoded by different allele types, such as HLA allele types, etc. For example, such training data may include information related to binning and selection of positive/negative MHC-peptide interaction. The training data may include one or more of positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data.

[0057] At 220 , a gradient descent process may be applied to the received data 218 for training. Gradient descent is an iterative process for performing machine learning, such as finding the minimum or local minimum of a function. For example, to find the minimum or local minimum of a function using gradient descent, the values of the variables are updated in increments proportional to the negative value of the gradient (or approximate gradient) of the function at the current point. For machine learning, the parameter space can be searched using gradient descent. Different gradient descent strategies can find different "destinations" in the parameter space, thus limiting the predicted errors to an acceptable degree. In embodiments, the gradient descent process can adapt the learning rate to the input parameters, for example, by performing large updates for rarely repeating parameters and smaller updates for frequently repeating parameters. Such embodiments may be suitable for dealing with sparse data. For example, a gradient descent strategy known as RMSprop can provide improved performance with peptide binding datasets.

[0058] At 221 , loss measurement may be applied to measure the loss or "cost" of processing. Examples of such loss measurements may include root mean square error or cross entropy.

[0059] At block 222 , it may be determined whether the exit criteria for gradient descent has been met. Because gradient descent is an iterative process, criteria can be specified for determining when the iterative process should stop by indicating that generator 228 is capable of generating positive simulated polypeptide-MHC-I interaction data that is classified by discriminator 226 as positive and/or real. At step 222 , if it is determined that the exit criteria for gradient descent has not been met, then the process may return to 220 , and the gradient descent process continues. At step 222 , if it is determined that the exit criteria for gradient descent has been met, the process may continue to step 224 , where discriminator 226 and generator 228 can be trained, for example, as described with reference to FIG. 5A . At 224 , the trained models for discriminator 226 and generator 228 may be stored. These stored models may include data defining the structure and coefficients that make up the models for discriminator 226 and generator 228 . Stored models provide the ability to use generator 228 to generate fake data and discriminator 226 to identify data, and with proper training, provide accurate and usable results from discriminator 226 and generator 228 .

[0060] The process may then continue from steps 230-238 which generally correspond to step 120 in FIG. 1 . In steps 230-238 , generated data instances (eg, positive simulated polypeptide-MHC-I interaction data) can be generated using a trained generator 228 . For example, at step 230 , the GAN generation process can be customized, for example, by setting a number of parameters 232 , 234 to control GAN generation 236 . Examples of parameters that may be set may include a generation size 232 and a sampling size 234 . Generate size options 232 may allow the size of the generated data set to be specified. For example, the size of the generated data set (positive simulated polypeptide-MHC-I interaction data) can be set to 2.5 times the size of the real data (positive real polypeptide-MHC-I interaction data and/or negative real polypeptide-MHC interaction data). -I). In this example, if the original real data in the batch is 64, then the corresponding generated simulated data in the batch is 160 . Sample size parameters 234 may provide the ability to specify the sample size that will be used to generate the data set. For example, this parameter can be given as a cutoff percentile for the 20 amino acid selection in the last layer of the generator. For example, specifying the 90th percentile means that all points below the 90th percentile will be set to 0, and the rest can be normalized using a normalization function such as the normalized exponential function (softmax). At step 236 , the trained generator 228 may be used to generate a dataset 236 that may be used to train the CNN model.

[0061] At step 240 , the simulated data instances 238 generated by the trained generator 228 and the real data instances from the original dataset may be mixed to form a new training dataset 240 , which generally corresponds to step 120 in FIG. 1 . The training data 240 may include one or more of positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data. In steps 242 - 262 , a convolutional neural network (CNN) classifier model 262 may be trained using the blended data 240 for training. At step 242 , the training of the CNN may be customized, for example, by setting a set of parameters 244 - 252 to control the training of the CNN 254 . Examples of parameters that can be set may include allele type 244 , allele length 246 , model complexity 248 , training rate 250 , and batch size 252 . The allele type parameters 244 may provide the ability to specify one or more allele types to be included in the CNN processing. Examples of these types of alleles are shown in Fig. 12 . For example, these alleles may include A0201, A0202, B2703, B2705, etc. shown in FIG. 12 . Allele length parameters 246 may provide the ability to specify the lengths of peptides that can bind to each specified allele type 244 . Examples of such allele lengths are shown in FIG. 13A . For example, for A0201 the specified length is displayed as 9 or 10, for A0202 the specified length is displayed as 9, for B2705 the specified length is displayed as 9, and so on. be used during training of 254 CNNs. Examples of such aspects may include the number of layers, the number of nodes per layer, the window size for each convolutional layer, etc. The learning rate parameters 250 may provide the ability to specify one or more rates at which the learning processing performed when training the CNN 254 should converge. . Examples of such learning rate parameters may include 0.001, which is a dimensionless parameter that defines the relative learning rate. Packet size parameters 252 may provide the ability to specify the sizes of training data packets 240 to be processed during CNN training 254 . For example, if the training dataset is divided into 100 equal parts, the batch size can be an integer of the training data size (train_data_size)/100. CNN training setting processing 242 may collect training parameters 244-252 , process them to be compatible with CNN training 254 , and input the processed parameters into CNN training 254 or store the processed parameters in appropriate files or locations for use in CNN training 254 .

[0062] At 254 , CNN training may begin. CNN training 254 may receive data 240 for training, for example, in bursts, as indicated by packet size parameters 252 . At 256 , a gradient descent process may be applied to the received data 240 for training. As described above, gradient descent is an iterative process for performing machine learning, such as finding the minimum or local minimum of a function. For example, a gradient descent strategy known as RMSprop can provide improved performance with peptide binding datasets.

[0063] At 257 , loss measurement can be used to measure the loss or "cost" of processing. Examples of such loss measurements may include root mean square error or cross entropy.

[0064] At block 258 , it may be determined whether the exit criteria for gradient descent has been met. Because gradient descent is an iterative process, criteria can be specified to determine when the iterative process should stop. At step 258 , if it is determined that the exit criteria for gradient descent has not been met, then the process may return to 256 , and the gradient descent process continues. At step 258 , if it is determined that the exit criteria for gradient descent is met (indicating that the gCNN is able to classify positive (real or simulated) polypeptide-MHC-I interaction data as positive and/or negative real polypeptide-MHC- I as negative), then the process may continue to 260 where the CNN classifier model 262 may be stored as the CNN classifier model 262 . These stored models may include data defining the structure and coefficients that make up the CNN classifier model 262 . The stored models provide the ability to use the CNN classifier model 262 to classify peptide bindings of input data instances and, if properly trained, provide accurate and usable results from the CNN classifier model 262 . At step 264 , CNN training ends.

[0065] In steps 266 - 280 , a trained convolutional neural network (CNN) classifier model 262 may be used to provide and evaluate predictions based on test data (the test data may comprise one or more of positive real-world polypeptide-MHC-I interaction data and/or negative real polypeptide-MHC-I interaction data) to measure the performance of the entire GAN model, which generally corresponds to step 130 in FIG. 1 . At step 270 , the GAN exit criteria may be set, for example, by setting a set of parameters 272 - 276 to control the evaluation process 266 . Examples of parameters that may be set may include prediction accuracy parameters 272 , prediction confidence parameters 274 , and loss parameters 276 . The prediction accuracy parameters 272 may provide the ability to determine the accuracy of the predictions to be provided by the score 266 . For example, the accuracy threshold for predicting a real positive category may be greater than or equal to 0.9. Prediction confidence parameters 274 may provide the ability to specify levels of confidence (eg, softmax normalization) for predictions to be provided by estimation 266 . For example, the fake or spurious category prediction confidence threshold may be set to a value greater than or equal to 0.4 and greater than or equal to 0.6 for the real negative category. The GAN output criteria tuning processing 270 may collect the training parameters 272-276 , process them so that they are compatible with the GAN predictive score 266 , and input the processed parameters into the GAN predictive score 266 , or store the processed parameters in appropriate files or locations for use in the predictive score. 266 GAN estimate. At 266 , a GAN predictive evaluation may be initiated. The GAN predictive score 266 may receive test data 268 .

[0066] At step 267 , a measurement of the area under the receiver operating characteristic curve (ROC) (AUC) may be performed. AUC is a normalized measure of classification efficiency. AUC measures the probability that, given two random points, one from the positive class and one from the negative class, the classifier will rate a point from a positive class higher than a point from a negative class. In fact, this is how ranking performance is measured. AUC is based on the idea that the more prediction classes that are mixed together (in the output space of the classifier), the worse the classifier performs. ROC scans the output space of a classifier with a moving boundary. At each scan point, false positives (FPR) and true positives (TPR) are recorded (as a normalized measure). The greater the difference between the two values, the less the points mix and the better they are classified. Once all FPR and TPR pairs have been obtained, they can be sorted and the ROC curve plotted. AUC is the area under this curve.

[0067] At step 278 , it can be determined whether the exit criteria for gradient descent has been met or not, which generally corresponds to step 140 in FIG. 1 Because gradient descent is an iterative process, criteria can be specified to determine when the iterative process should stop. At step 278 , if it is determined that the exit criteria for the evaluation process 266 were not met, then the process may return to 220 , and the GAN training process 220-264 and the evaluation process 266 continue. Thus, when the exit criterion is not met, the process will return to GAN training (generally corresponding to returning to step 110 in FIG. 1 ) to create a generator with better results. At step 278 , if it is determined that the exit criteria from evaluation process 266 are met (indicating that CNN has classified positive real polypeptide-MHC-I interaction data as positive and/or negative real polypeptide-MHC-I interaction data as negative) , then the process may continue to 280 , where the predictive evaluation processing and process 200 are terminated, which generally corresponds to step 150 in FIG. 1 .

[0068] An exemplary embodiment of the internal processing structure of generator 228 is shown in FIG. 6 - fig. 7 . In this example, each processing unit may perform the specified type of processing and may be performed in the order shown. It should be noted that this is just an example. In embodiments, the types of processing performed, as well as the order in which processing is performed, may be changed.

[0069] FIG. 6 - fig. 7 describes an example processing flow for generator 228 . The processing flow is only an example and is not intended to be limiting. The processing included in generator 228 may begin with density processing 602 , in which input data is fed into a feed-forward neural layer to estimate the spatial variation in input data density. At 604 , batch normalization processing may be performed. For example, normalization processing may include adjusting values measured at different scales to a common scale to adjust all probability distributions of data values for matching. This normalization can provide an increased rate of convergence because the original (deep) neural networks are sensitive to changes in the layers at the beginning, and the direction parameter is optimized to be distracted by trying to reduce errors for outliers in the data at the beginning. Batch normalization arranges the gradients away from these distractions and is therefore faster. At 606 , activation processing may be performed. For example, activation processing may include tanh, a sigmoid function, ReLU (rectified linear units), or a step function, etc. For example, ReLU has an output of 0 if the input is less than 0, and a raw input otherwise. It is simpler (requires less computation) compared to other activation functions and thus can provide faster learning. At block 608 , input reshape processing may be performed. For example, such processing can help convert the input shape(s) to a target shape that can be accepted as valid input in the next step. At 610 , Gaussian elimination processing may be performed. Dropout is a regularization technique to reduce overfitting in neural networks based on certain training data. Elimination can be made by removing neural network nodes that may cause or exacerbate overfitting. Gaussian elimination processing may use a Gaussian distribution to determine the nodes to be removed. Such processing may produce noise as elimination, but may store the mean and variance of the input data relative to their original values based on a Gaussian distribution to ensure the self-normalization property even after elimination.

[0070] At 612 , Gaussian noise processing may be performed. Gaussian noise is statistical noise having a probability density function (PDF) equal to a normal, or Gaussian, distribution function. Gaussian noise processing can involve adding noise to the data to prevent the model from learning small (often trivial) changes in the data, thereby adding resilience to model overfitting. This process can improve the accuracy of forecasts. At block 614 , two-dimensional (2D) convolutional processing may be performed. 2D convolution is an extension of 1D convolution by convolving both the horizontal and vertical directions in a two-dimensional spatial domain and can provide data smoothing. Such processing can scan all partial inputs with multiple moving filters. Each filter can be thought of as a neural layer with shared parameters that counts the occurrence of a particular feature (matching the values of the filter's parameters) at all locations in the feature map. At 616 , a second batch normalization processing may be performed. At block 618 , a second activation processing may be performed, at block 620 , a second Gaussian elimination processing may be performed, and at block 622 , 2D upsampling processing may be performed. Upsampling processing can convert the input data from its original form to the desired (mostly larger) form. For example, resampling or interpolation can be used for this. For example, the input data can be rescaled to the desired size, and the value at each point can be calculated using an interpolation such as bilinear interpolation. At block 624 , a second Gaussian noise processing may be performed, and at block 626 , two-dimensional (2D) convolution processing may be performed.

[0071] In FIG. 7 , a third batch normalization processing may be performed at 628 , a third activation processing may be performed at 630 , a third Gaussian elimination processing may be performed at 632 , and a third Gaussian noise processing may be performed at 634 . At block 636 , a second two-dimensional (2D) convolutional processing may be performed, and at block 638 , a fourth batch normalization processing may be performed. Activation processing may be performed after step 638 and before step 640 . At 640 , a fourth Gaussian elimination processing may be performed.

[0072] At block 642 , a fourth Gaussian noise processing may be performed, at block 644 , a third two-dimensional (2D) convolutional processing, and at block 646 , a fifth batch normalization processing may be performed. At 648 , a fifth Gaussian elimination processing may be performed, at 650 , a fifth Gaussian noise processing may be performed, and at 652 , a fourth activation processing may be performed. Such activation processing can use a sigmoid activation function that converts the input from [- infinity, infinity] to the output [0,1]. Typical data recognition systems may use an activation function on the last layer. However, due to the categorical nature of the present techniques, the sigmoid function may provide improved prediction of MHC binding. The sigmoid function is more powerful than ReLU and can provide a suitable probability output. For example, in a present classification problem, inference as a probability may be desirable. However, since the sigmoid function can be much slower than ReLU or tanh, for efficiency reasons it may not be desirable to use the sigmoid function for previous activation layers. However, since the last dense layers are more directly related to the final inference, applying the sigmoid function on this activation layer can significantly improve convergence compared to ReLU.

[0073] At block 654 , second input reshaping processing may be performed to change the output relative to the dimension of the data (which may be fed to the discriminator later).

[0074] An exemplary embodiment of the discriminator 226 processing flow is shown in FIG. 8 - fig. 9 . The processing flow is only an example and is not intended to be limiting. In this example, each processing unit may perform the specified type of processing and may be performed in the order shown. It should be noted that this is just an example. In embodiments, the types of processing performed, as well as the order in which processing is performed, may be changed.

[0075] Returning to FIG. 8 , the processing included in the discriminator 226 may begin with a one-dimensional (1D) convolutional processing 802 that may take an input, apply a 1D convolution filter to the input, and produce an output. At 804 , batch normalization processing may be performed, and at 806 , activation processing may be performed. For example, Leaky Rectified Linear Unit (RELU) processing may be used to perform activation processing. RELU is one type of activation function for a node or neuron in a neural network. A leaky RELU can tolerate a small non-zero gradient when the node is not active (input less than 0). ReLU has a problem called "dying" where it keeps outputting 0 when the activation function's input has a large negative bias. When this happens, the model stops training. Leaky ReLU solves this problem by providing a non-zero gradient even when it is inactive. For example, f(x)=alpha*x for x<0, f(x)=x for x>=0. At block 808 , input reshaping processing may be performed, and at block 810 , 2D upsampling processing may be performed.

[0076] Optionally, at block 812 , Gaussian noise processing may be performed, at block 814 , two-dimensional (2D) convolution processing may be performed, at block 816 , a second batch normalization processing may be performed, at block 818 , a second activation processing may be performed, at block 820 , second 2D upsampling processing, a second 2D convolutional processing may be performed at 822 , a third batch normalization processing may be performed at 824 , and a third activation processing may be performed at 826 .

[0077] Continuing with FIG. 9 ; at step 828 a third 2D convolution processing may be performed, at step 830 a fourth batch normalization processing may be performed, at step 832 a fourth activation processing may be performed, at step 834 a fourth 2D convolution processing may be performed, at step 836 a fifth batch normalization processing may be performed , at step 838 , a fifth activation processing may be performed, and at step 840 , data alignment processing may be performed. For example, data flattening processing may include combining data from different tables or datasets to form a single table or dataset, or a reduced number thereof. At 842 , density processing may be performed. At block 844 , a sixth activation processing may be performed, at block 846 , a second density processing may be performed, at block 848 , a sixth batch normalization processing may be performed, and at block 850 , a seventh activation processing may be performed.

[0078] The sigmoid function can be used instead of the leaky ReLU as the activation functions for the last 2 dense layers. The sigmoid is more powerful than a leaky ReLU and can provide an output with a reasonable probability (for example, in a classification problem, an output as a probability is desirable). However, the sigmoid function is slower than the leaky ReLU, the use of the sigmoid may not be desirable for all layers. However, since the last two dense layers are more directly related to the final output, sigmoid improves convergence significantly compared to leaky ReLU. In embodiments, two dense layers (or fully connected neural network layers) 842 and 846 can be used to obtain enough complexity to transform their inputs. In particular, a single dense layer may not be complex enough to transform the results of the convolution into the output space of the discriminator, although this may be sufficient for use in a generator 228 .

[0079] In an embodiment, methods are disclosed for using a neural network (eg, CNN) to classify input data based on a previous learning process. The neural network may generate a prediction score and thus may classify the input biological data as pass or fail based on the neural network pre-trained on the set of pass and fail biological data, including the prediction scores. The prediction scores may be binding affinity scores. A neural network can be used to generate a predictive binding affinity score. The binding affinity index can numerically represent the probability that one biomolecule (eg, protein, DNA, drug, etc.) will bind to another biomolecule (eg, protein, DNA, drug, etc.). The predicted binding affinity score can numerically represent the likelihood that a peptide (eg, MHC) will bind to another peptide. However, until now, machine learning techniques could not be applied due to at least the inability to produce reliable predictions when a neural network is trained on small amounts of data.

[0080] The described methods and systems solve this problem by applying a combination of features for more reliable prediction. The first feature is the use of an extended set of biological training data to train a neural network. This advanced training set is developed by teaching GANs to generate simulated biological data. The neural networks are then trained with this augmented training set (eg, using stochastic backpropagation learning, which is a type of machine learning algorithm that uses the gradient of a mathematical loss function to adjust the network weights). Unfortunately, the introduction of an augmented training set can increase the number of false positives in the classification of biological data. Accordingly, the second feature of the described methods and systems is to minimize these false positives by executing an iterative learning algorithm as needed, in which the GAN is further employed to generate an updated training simulation set containing higher quality simulation data, and the neural network is retrained by the updated training set. for learning. This combination of features provides a robust predictive model that can predict the success (eg, binding affinity scores) of certain biological data, limiting false positives.

[0081] The dataset may contain unclassified biological data, such as unclassified protein interaction data. Unclassified biological data may include data relating to a protein for which an estimate of binding affinity for another protein is not available. The biological data may include a plurality of candidate protein-protein interactions, such as candidate protein-MHC-I interaction data. The CNN may generate a prediction score indicating binding affinity and/or classify each of the candidate polypeptide-MHC-I interactions as positive or negative.

[0082] In the embodiment shown in FIG. 10 , a computer-implemented method 1000 for training a neural network to predict binding affinity may include collecting a set of positive biological data and negative biological data from a database at 1010 . The biological data may include protein-protein interaction data. The protein-protein interaction data may include one or more of the following: a first protein sequence, a second protein sequence, a first protein identifier, a second protein identifier and/or a binding affinity score, and the like. In one embodiment, the binding affinity score may be equal to 1, indicating successful binding (eg, positive biological data), or -1, indicating unsuccessful binding (eg, negative biological data).

[0083] A computer-implemented method 1000 may include applying a generative adversarial network (GAN) to a set of positive biological data to create a set of simulated positive biological data at 1020 . Applying a GAN to a set of positive biological data to generate a set of simulated positive biological data may include generating increasingly accurate positive simulated biological data by the GAN generator until the GAN discriminator classifies the positive simulated biological data as positive.

[0084] The computer-implemented method 1000 may include generating a first training set containing a collected positive biological data set, a simulated positive biological data set, and a negative biological data set at 1030 .

[0085] A computer-implemented method 1000 may include training a neural network in a first stage using a first training set in stage 1040 . Training a neural network in the first stage using the first training set may involve submitting positive simulated biological data, positive biological data, and negative biological data to a convolutional neural network (CNN) until the CNN is able to classify the biological data as positive or negative. .

[0086] The computer-implemented method 1000 may include generating a second training set for the second stage of training by reapplying the GAN to generate additional simulated positive biological data in stage 1050 . The creation of a second training set may be based on presenting positive biological data and negative biological data to the CNN to generate prediction scores and determine that the prediction scores are inaccurate. The prediction scores may be binding affinity scores. Inaccurate prediction scores indicate that the CNN is not fully trained, which can be traced back to finding that the GAN is not fully trained. Accordingly, one or more iterations of the GAN generator may be performed generating progressively more accurate positive simulated biological data until the GAN discriminator classifies the positive simulated biological data as positive to generate additional simulated positive biological data. The second training set may comprise positive biological data, simulated positive biological data, and negative biological data.

[0087] A computer-implemented method 1000 may include training a neural network in a second stage using a second training set in stage 1060 . Training the neural network in the second stage using the second training set may involve submitting positive biological data, simulated positive biological data, and negative biological data to the CNN until the CNN is able to classify the biological data as positive or negative.

[0088] After the CNN is fully trained, new biological data can be presented to the CNN. New biological data may include data on protein-protein interactions. The protein-protein interaction data may include one or more of a first protein sequence, a second protein sequence, a first protein identifier and/or a second protein identifier, and the like. The CNN may analyze the new biological data and generate a prediction score (e.g., predicted binding affinity) , indicating the predicted successful or unsuccessful binding.

[0089] In an illustrative aspect, the methods and systems may be implemented on computer 1101 as illustrated in FIG. 11 and described below. Likewise, the disclosed methods and systems may use one or more computers to perform one or more functions at one or more locations. In FIG. 11 is a block diagram illustrating an exemplary operating environment for performing the disclosed methods. This exemplary operating environment is only an example of an operating environment and is not intended to suggest any limitation as to the scope of application or functionality of the operating environment architecture. Also, the operating environment should not be construed as in any way dependent on or requiring the presence of any of the components, or combinations thereof, shown in the illustrative operating environment.

[0090] The methods and systems of the present invention may operate with numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and/or configurations that may be suitable for use with the systems and methods include, without limitation, personal computers, server computers, laptop computers, and multiprocessor systems. Additional examples include set-top boxes, programmable consumer electronics, networked computers, minicomputers, supercomputers, distributed computing environments that include any of the above systems or devices, and the like.

[0091] The computation provided by the disclosed methods and systems can be performed using software components. The disclosed systems and methods may be described in the general context of computer-executable instructions, such as blocks of programs executed by one or more computers or other devices. Generally, blocks of programs include program code, algorithms, programs, objects, components, data structures, etc., that perform specific tasks or provide implementations of specific abstract data types. The disclosed methods may also be practiced using networked and distributed computing environments in which tasks are performed by remote data processing devices that are linked through a communications network. In a distributed computing environment, program blocks can be located both in local and remote computer storage media, including storage devices.

[0092] In addition, one of skill in the art will appreciate that the systems and methods disclosed herein may be implemented by a general purpose computing device in the form of computer 1101 '. Computer 1101 components may include, without limitation, one or more processors 1103 , system storage 1112 , and a system bus 1113 that connects various system components, including one or more processors 1103 to system storage 1112 . The system may use parallel computing.

[0093] The system bus 1113 is one or more of several possible types of bus structures, including a storage bus or storage controller, a peripheral bus, an accelerated graphics port, or a local bus using any of a variety of bus architectures. By way of example, such architectures may include an industry standard architecture (ISA) bus, a microchannel architecture (MCA) bus, an enhanced industry standard architecture (EISA) bus, a Video Electronics Standards Association (VESA) local bus, an AGP extension bus for Video Adapter Connections (AGP) and Local Peripheral Interconnect (PCI), Serial/PCI Express Serial, Personal Computer International Association (PCMCIA) Memory Card, Universal Serial Bus (USB), etc. 1113 bus and all buses described herein may also be implemented via a wired or wireless network connection, wherein each of the subsystems, including one or more processors 1103 , mass storage 1104 , operating system 1105 , classification software 1106 (e.g., GAN , CNN), classification data 1107 (e.g. "real e" or "simulated" data, including positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and/or negative real polypeptide-MHC-I interaction data), network adapter 1108 , system storage device 1112 , I/O interface 1110 , graphics adapter 1109 , display device 1111 , and user-machine interface 1102 may reside within one or more remote computing devices 1114a,b,c located in physically separate locations connected via buses of this type, resulting in a fully distributed system.

[0094] Computer 1101 typically includes a number of different computer-readable media. Exemplary storage media can be any available media accessed by computer 1101 and includes, for example, without limitation, both volatile and nonvolatile media, removable and non-removable media. System storage 1112 includes computer-readable media in the form of volatile storage, such as random access memory (RAM), and/or non-volatile storage, such as read only memory (ROM). System memory 1112 typically contains data, such as classification data 1107 , and/or blocks of programs, such as operating system 1105 and classification software 1106 , that are directly accessible to one or more processors 1103 and/or are currently being processed with their help.

[0095] In another aspect, computer 1101 may also include other removable/non-removable volatile/non-volatile computer storage media. As an example, in FIG. 11 shows a mass storage device 1104 that can provide non-volatile storage of program code, computer readable instructions, data structures, program blocks, and other data for computer 1101 . For example, without limitation, the mass storage device 1104 may be a hard drive, removable magnetic disk, removable optical disk, magnetic cassettes or other magnetic storage devices, flash memory cards, CD-ROMs, digital versatile disks (DVDs), or other optical storage media. device, random access memory (RAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), etc.

[0096] Optionally, any number of program modules may be stored on mass storage device 1104 , including by way of example an operating system 1105 and classification software 1106 . Both the operating system 1105 and the classification software 1106 (or any combination thereof) may contain program elements and the classification software 1106 . Classification data 1107 may also be stored on mass storage device 1104 . Classification data 1107 may be stored in any one or more databases known in the art. Examples of such databases include DB2®, Microsoft® Access, Microsoft® SQL Server, Oracle®, mySQL, PostgreSQL, etc. Databases can be centralized or distributed across multiple systems.

[0097] In another aspect, a user may enter commands and information into the computer 1101 using an input device (not shown). Examples of such input devices include, without limitation, a keyboard, a pointing device ( e.g. "mouse"), a microphone, a joystick, a scanner, tactile input devices such as gloves and other body coverings, and the like. These and other input devices data may be connected to one or more processors 1103 via a user-machine interface 1102 that is connected to the system bus 1113 , but may be connected via another interface and bus structures such as a parallel port, a game port, an IEEE 1394 port (also known as a Firewire port), serial port, or Universal Serial Bus (USB).

[0098] In yet another aspect, the display device 1111 can also be connected to the system bus 1113 via an interface, such as a graphics adapter 1109 '. It is contemplated that computer 1101 may have more than one graphics adapter 1109 , and computer 1101 may have more than one display device 1111 . For example, the display device 1111 may be a monitor, an LCD (liquid crystal display), or a projector. In addition to display device 1111 , other peripheral output devices may include components such as speakers (not shown) and a printer (not shown) that may be connected to computer 1101 via I/O interface 1110 . Any step and/or result of the methods may be output in any form to an output device. Such output may be in any form of visual presentation, including, without limitation, textual, graphical, animated, audio, tactile, and the like. Display device 1111 and computer 1101 may be part of the same device or separate devices.

[0099] The computer 1101 may operate in a networked environment using logical connections to one or more remote computing devices 1114a, b, c . As an example, the remote computing device may be a personal computer, laptop, smartphone, server, router, network computer, peer device, or other common network node, etc. Logical connections between the computer 1101 and the remote computing device 1114a, b, c may be established via a network 1115 such as a local area network (LAN) and/or a shared wide area network (WAN). Such network connections can be made through the network adapter 1108 . Network adapter 1108 can be implemented in both wired and wireless environments. Such networking environments are traditional and commonplace in homes, offices, enterprise-wide computer networks, intranets, and the Internet.

[00100] For purposes of illustration, application programs and other executable program components, such as the operating system 1105 , are illustrated herein as discrete units, although it is understood that such programs and components reside at different times in different memory components of a computing device. 1101 and their execution is implemented by one or more processors 1103 of the computer. An implementation of the classification software 1106 may be stored on computer-readable media or transmitted via some type of computer-readable media. Any of the disclosed methods may be carried out by computer-readable instructions implemented on computer-readable media. Computer-readable media can be any available media that can be accessed by a computer. By way of example and without limitation, computer-readable media may include "computer storage media" and "communications media". "Computer storage media" includes volatile and nonvolatile removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules, or other data. Illustrative computer storage media include, without limitation, RAM, ROM, EEPROM, flash memory or other information storage technology, CD-ROM, digital versatile disks (DVD) or other optical storage device, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other media that can be used to store the required information and that can be accessed using a computer.

[00101] The methods and systems may use artificial intelligence techniques such as machine learning and iterative learning. Examples of such techniques include, without limitation, expert systems, case-based reasoning, Bayesian networks, behavioral artificial intelligence, neural networks, fuzzy logic systems, evolutionary modeling (e.g., genetic algorithms), swarm intelligence (e.g., ant colony algorithms), and hybrid intelligent systems (for example, expert inference rules obtained through a neural network, or production rules obtained as a result of statistical learning).

[00102] The following examples are set forth to provide those skilled in the art with a complete disclosure and description of how the compounds, compositions, articles, devices, and/or methods claimed herein are prepared and evaluated, and are intended to be are illustrative only and are not intended to limit the scope of the methods and systems. Efforts have been made to ensure accuracy in terms of numbers (eg quantities, temperatures, etc.), but some errors and deviations must be taken into account. Unless otherwise indicated, parts are parts by weight, temperatures are in °C or provided at ambient temperature, and pressures are equal to or close to atmospheric.

B. HLA alleles

[00103] The disclosed systems can be trained on an unlimited number of HLA alleles. Data on peptide binding to MHC-I protein complexes encoded by HLA alleles are known in the art and are available from databases including, but not limited to, IEDB, AntiJen, MHCBN, SYFPEITHI, and the like.

[00104] In one embodiment, the disclosed systems and methods improve the predictive value of peptide binding to MHC-I protein complexes encoded by HLA alleles: A0201, A0202, B0702, B2703, B2705, B5701, A0203, A0206, A6802, and combinations thereof. For example, 1028790 is a test case for A0201, A0202, A0203, A0206, A6802.

allele test set A0201 1028790 A0202 1028790 B0702 1028928 B2703 315174 B2705 1029125 B5701 1029061 A0203 1028790 A0206 1028790 A6802 1028790

[00105] Predictability can be improved over existing neural systems, including but not limited to NetMHCpan, MHCflurry, sNeubula, and PSSM.

III. Therapeutic agents

[00106] The disclosed systems and methods are useful for identifying peptides that bind to MHC-I of T cells and target cells. In one embodiment, the peptides are tumor-specific peptides, viral peptides, or a peptide that is displayed on the MHC-I of the target cell. The target cell may be a tumor cell, a cancer cell, or a cell infected with a virus. Peptides are typically displayed on antigen-presenting cells, which then present the peptide antigen to CD8+ cells, such as cytotoxic T cells. Binding of a peptide antigen to a T cell activates or stimulates the T cell. Thus, in one embodiment, a vaccine, such as a cancer vaccine, is provided, comprising one or more of the peptides identified by the disclosed systems and methods.

[00107] Another embodiment relates to an antibody, or antigen-binding fragment thereof, that binds to a peptide, a peptide antigen-MHC-I complex, or both.

[00108] While specific embodiments of the present invention have been described, those skilled in the art will appreciate that there are other embodiments that are equivalent to the described embodiments. Accordingly, it should be understood that the present invention is not limited to the specific illustrated embodiments, but only by the scope of the appended claims.

EXAMPLES

Example 1: Evaluation of Existing Predictive Models

[00109] NetMHCpan, sNebula, MHCflurry, CNN, PSSM prediction models were evaluated. The area under the ROC curve was used as a measure of effectiveness. A value of 1 is good performance, 0 is poor performance, and 0.5 is a random guess. Table 1 shows the models and data used.

Table 1: Various models for predicting peptide binding to MHC-I protein complexes encoded by the indicated alleles

NetMHCpan Paired neural network sNebula SVM designed to find pairwise similarity MHCflurry Ensemble of Neural Networks CNN Convolutional Neural Network PSSM Positional weight matrix

[00110] FIG. 12 shows evaluation data showing that a CNN trained as described here outperforms other models in most test cases, including the current counterpart, NetMHCpan. In FIG. 12 is an AUC heatmap showing the results of applying the current models and currently described methods ("CNN_ours") to the same 15 test datasets. In FIG. 12 diagonal lines from bottom left to top right usually indicate a higher value, the thinner the lines, the higher the value, and the thicker the lines, the lower the value. Diagonal lines from bottom right to top left indicate a lower value, the thinner the lines, the lower the value, and the thicker the lines, the higher the value.

Example 2: CNN Model Problems

[00111] CNN training involves many random processes (eg, mini-batch feeds, stochastic element entering gradient by elimination, noise, etc.), so reproducibility of the training process can be problematic. For example, in FIG. 12 shows that AUC Vang("Yeeling") cannot be reproduced perfectly when implementing the same algorithm on the same data. Vang, et al., HLA class I binding prediction via convolutional neural networks, Bioinformatics, Sep 1;33(17):2658-2665 (2017).

[00112] In general, CNN is less complex than other deep learning frameworks such as the Deep Neural Network due to the nature of parameter sharing, however it is still a complex algorithm.

[00113] The standard CNN provides feature extraction from data using a fixed size window, but the peptide binding information may not be encoded with the same length. In the present disclosure, since biology studies have shown that one type of binding mechanism occurs at the scale of 7 amino acids in the peptide chain, a window size of 7 can be used, and although this window size provides good performance, it may not be sufficient to explain other types. binding factors in all HLA binding problems.

[00114] FIG. 13A-FIG. 13C shows the discrepancy between different models. In FIG. 13A shows 15 test data sets from the HLA binding data released weekly by the IEDB. Test_id is marked by the authors of this disclosure as a unique identifier for all 15 test cases. IEDB is the IEDB data release identifier, there can be several different subsets of data in one IEDB release belonging to different HLA categories. HLA is a type of HLA that binds to peptides. The length is the binding length of the peptides to HLA. The test size is the number of records that this test set has. The training size is the number of records in this training set. The ratio of bindings is the ratio of bindings to the sum of bindings and the absence of bindings in the training dataset, the authors list it in this document to measure the skewness of the training data. The size of bindings is the number of bindings in the training dataset, the authors use it to calculate the percentage of bindings.

[00115] FIG. 13B-FIG. 13C demonstrates the difficulty in reproducing a CNN implementation. As regards the differences between the models, in Fig. 13B-FIG. 13C shows 0 differences between models. In FIG. 13B -FIG. 13C shows that Adam's implementation does not match the published results.

Example 3: Offset in datasets

[00116] A training/test set split was performed. The training/test set split is a measurement designed to prevent overfitting, however the effectiveness of the measurement may depend on the data selected. Performance varies significantly between models regardless of how they are tested on the same MHC allele (A*02:01). This shows the AUC bias obtained by selecting the biased test set, FIG. 14 . The results using the described methods on the biased training/test set are indicated in the "CNN*1" column, which shows lower performance than that shown in FIG. 12 . In FIG. 14 diagonal lines from bottom left to top right usually indicate a higher value, the thinner the lines, the higher the value, and the thicker the lines, the lower the value. Diagonal lines from bottom right to top left indicate a lower value, the thinner the lines, the lower the value, and the thicker the lines, the higher the value.

Example 4: SRCC Offset

[00117] The best Spearman Rank Correlation Coefficient (SRCC) was selected from the 5 models tested and compared to the size of the normalized data. In FIG. 15 shows that the smaller the test size, the better the SRRC. SRCC measures the discrepancy between prediction rank and label rank. The larger the test size, the more likely it is that the ranking order will be violated.

Example 5: Gradient Descent Comparison

[00118] A comparison was made between Adam and RMSprop. Adam is a first order gradient optimization algorithm for stochastic objective functions based on lower order adaptive moment estimates. RMSprop (for root mean square propagation) is also the way in which the learning rate is adapted for each of the parameters.

[00119] FIG. 16A - FIG. 16C shows that RMSprop has an improvement over most of the dataset compared to Adam. Adam is a momentum-based optimizer that changes the parameters significantly in the beginning compared to RMSprop. The improvement may refer to the following: 1) since the discriminator is in charge of the entire training process of the GAN, if it follows the pulse and updates its parameters significantly, then the generator terminates in a non-optimal state; 2) peptide data is different from images, which allow fewer errors when generated. A small difference of 9~30 positions can significantly change the linking results, while whole image pixels can be changed but remain in the same image category. Adam is committed to further exploration in the options zone, but that means it's easier for each position in the zone; while RMSprop pauses longer at each point and can detect small parameter changes indicating a significant improvement in the end result of the discriminator and pass that knowledge on to the generator to generate better modeled peptides.

Example 5: Peptide Training Format

[00120] Table 2 shows an example of MHC-I interaction data. Shown are peptides with different binding affinities for the indicated HLA allele. The peptides were designated as binding (1) or non-binding (-1). The binding category was converted from the half-maximal inhibitory concentration (IC ₅₀ ). The predicted output is given in units of IC ₅₀ nM. A lower number indicates a higher affinity. Peptides with IC ₅₀ values <50 nM are considered high affinity, those with <500 nM intermediate affinity, and <5000 nM low affinity. Most of the known epitopes are characterized by high or intermediate affinity. Some are characterized by low activity. None of the known T cell epitopes has an IC ₅₀ value greater than 5000 nM.

Table 2. Peptides for an identified HLA allele showing peptide binding or not binding to the MHC-I protein complex encoded by the HLA allele.

Peptide HLA Binding Category AAAAAAAALY (SEQ ID NO:1) A829:02 one AAAAALQAK (SEQ ID NO:2) A*03:01 one AAAAALWL (SEQ ID NO:3) C*16:01 one AAAAARAAL (SEQ ID NO:4) B*14:02 -one AAAAEEEEE (SEQ ID NO:5) A*02:01 -one AAAAFEAAL (SEQ ID NO:6) B*48:01 one AAAAPYAGW (SEQ ID NO:7) B*58:01 one AAAARAAAL (SEQ ID NO:8) B*14:02 one AAAATCALV (SEQ ID NO:9) A*02:01 one AAAATCALV (SEQ ID NO:9) A*02:02 one AAAATCALV (SEQ ID NO:9) A*02:03 one AAAATCALV (SEQ ID NO:9) A*02:06 one AAAATCALV (SEQ ID NO:9) A*68:02 one AAADAAAAL (SEQ ID NO:10) C*03:04 one AAADFAHAE (SEQ ID NO:11) B*44:03 -one AAADPKVAF (SEQ ID NO:12) C*16:01 one

Example 6: Comparing GANs

[00121] FIG. 17 shows that a combination of simulated (eg, fake, fake) positive data, real positive data, and real negative data results in a better prediction than real positive and real negative data alone, or simulated positive data and real negative data. The results of the methods described are shown in the "CNN" column and the two "GAN-CNN" columns. In FIG. 17 diagonal lines from bottom left to top right usually indicate a higher value, the thinner the lines, the higher the value, and the thicker the lines, the lower the value. Diagonal lines from bottom right to top left indicate a lower value, the thinner the lines, the lower the value, and the thicker the lines, the higher the value. GAN improves the performance of A0201 on all test sets. Using an information extractor (eg, CNN+skip-gram embedding) works well for peptide data because the binding information is spatially encoded. The data obtained from the exposed GAN can be thought of as a way of "conditional calculus" that helps make the distribution of the data smoother, making it easier to train the model. In addition, the GAN loss function forces the GAN to create clear instances rather than blue mean, which is different from classical ways such as variational autoencoders. Since there are many potential chemical bonding patterns, averaging different patterns towards the midpoint would be optimal, hence even if the GAN can overfit and face the mode collapse problem, it will model the patterns better.

[00122] The disclosed methods are superior to current systems in part because of the use of different training data. The disclosed methods are superior to using only real positive and real negative data because the generator can increase the frequency of some weak binding signals, which increases the frequency of some binding patterns, and balances the weights of different binding patterns in the training dataset, which makes model training easier.

[00123] The disclosed methods are superior to using only fake positives and real negatives because the fake positive data class has a mode collapse problem, meaning it cannot represent binding patterns for the entire population; similar to entering real positive and real negative data into the model as training data, but this reduces the number of training instances, resulting in the model having less data to use in training.

[00124] FIG. 17 , the following columns are used: test identifier: unique for one test case, used to distinguish between test cases; IEDB: data set identifier in the IEDB database; HLA: peptide binding complex allele type; length: number of amino acids in peptides; test size: number of observations found in this test dataset; training size: the number of observations in this training dataset; proportion of associations: percentage of associations in the training dataset; binding size: the number of bindings in the training dataset.

[00125] Unless expressly stated otherwise, it is by no means intended that any method set forth herein should be construed as requiring its steps to be performed in a particular order. Accordingly, if the claims do not actually specify the order of the steps, or the claims or the description do not specifically state that the steps are to be limited to a particular order, then the order is not intended to be thus taken out of context. This is true for any possible implicit basis for interpretation, including the logic of the order of stages or the sequence of execution of technological operations; commonly used meaning derived from grammatical construction or punctuation; the number or type of embodiments described in the specification.

[00126] Although in the foregoing description the present invention has been described in relation to some of its embodiments and many details have been given for the purpose of illustration, it will be apparent to those skilled in the art that the present invention is capable of additional embodiments and that some of the details described in this document, may vary significantly without deviating from the basic principles of the present invention.

[00127] All references cited herein are incorporated by reference in their entirety. The present invention may be embodied in other specific forms without deviating from its spirit or essential attributes, and accordingly, reference should be made to the appended claims rather than to the foregoing description, which indicates the scope of the present invention.

Illustrative Embodiments

Embodiment 1. A method for training a generative adversarial network (GAN) comprising: generating, by means of a GAN generator, progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive; presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC interaction data -I as positive or negative; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN to generate predictive scores; determining, based on the prediction estimates, that the GAN is trained; and derivation of GANs and CNNs.

Embodiment 2. The method of Embodiment 1, wherein generating increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as real, comprises: generating via a generator GAN according to the GAN parameter set of the first simulated dataset containing simulated positive polypeptide-MHC-I interactions for the MHC allele; combining the first simulated data set with positive real polypeptide-MHC-I interactions for the MHC allele and negative real polypeptide-MHC-I interactions for the MHC allele to create a GAN training data set; determining, with a discriminator according to a decision boundary, whether the polypeptide-MHC-I interaction for the MHC allele in the GAN training dataset is simulated positive, real positive, or real negative; an adjustment based on the accuracy of the discriminator's determination of one or more of the GAN parameter set or decision boundary; and repeating steps a-d until the first stop criterion is met.

Embodiment 3. The method of Embodiment 2, wherein presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, provides for: generating, by means of a GAN generator in accordance with the GAN parameter set, a second simulated dataset containing simulated positive polypeptide-MHC-I interactions for the HLA allele; combining a second simulated dataset, positive real-life polypeptide-MHC-I interactions for the MHC allele, and negative real-life polypeptide-MHC-I interactions for the MHC allele to create a CNN training dataset; representing a CNN training dataset into a convolutional neural network (CNN); classifying by the CNN according to the polypeptide-MHC-I interaction CNN parameter set for the MHC allele in the CNN training data set as positive or negative; an adjustment based on the classification accuracy by the CNN, one or more of the set of CNN parameters; and repeating steps h-j until the second stopping criterion is met.

Embodiment 4. The method of Embodiment 3, wherein presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in a CNN to generate predictive scores comprises: classifying by the CNN according to a set of polypeptide interaction CNN parameters -MHC-I for MHC allele as positive or negative.

Embodiment 5. The method of Embodiment 4, wherein determining based on the prediction scores that the GAN has been trained comprises determining the classification accuracy by the CNN, wherein when (if) the classification accuracy satisfies the third stopping criterion, deriving the GAN and the CNN.

Embodiment 6. The method of Embodiment 4, wherein determining based on the prediction scores that the GAN has been trained comprises determining the classification accuracy by the CNN, wherein when (if) the classification accuracy does not satisfy the third stopping criterion, returning to step a.

Embodiment 7. The method of Embodiment 2, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 8. The method of Embodiment 2 wherein the MHC allele is an HLA allele.

Embodiment 9. The method of Embodiment 8 wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 10. The method of Embodiment 8 wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 11. The method of Embodiment 8 wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 12. The method of Embodiment 1, further comprising: presenting a data set to a CNN, where the data set contains a plurality of candidate polypeptide-MHC-I interactions; classifying by CNN each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified as a positive polypeptide-MHC-I interaction.

Embodiment 13 A polypeptide obtained by the method of Embodiment 12.

Embodiment 14. The method of Embodiment 12, wherein the polypeptide is a tumor-specific antigen.

Embodiment 15. The method of Embodiment 12, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele.

Embodiment 16. The method of Embodiment 1, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with the selected allele.

Embodiment 17. The method of Embodiment 16 wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 18. The method of Embodiment 1, wherein generating increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive, involves evaluating a gradient descent expression. for the GAN generator.

Embodiment 19. The method of Embodiment 1, wherein generating increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive, involves: iteratively performing ( for example, optimizing) the GAN discriminator to increase the ability to assign high probability to positive real polypeptide-MHC-I interaction data, low probability to positive simulated polypeptide-MHC-I interaction data, and low probability to negative real polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the likelihood that positive simulated polypeptide-MHC-I interaction data will score high.

Embodiment 20. The method of Embodiment 1, wherein presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, provides for: performing a convolution procedure; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 21. The method of Embodiment 1, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 22. The method of Embodiment 2, wherein the first stopping criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 23. The method of Embodiment 3, wherein the second stopping criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 24. The method of Embodiment 5 or 6, wherein the third stopping criterion includes an estimate of an area under the curve (AUC) function.

Embodiment 25. The method of Embodiment 1, wherein the predictive score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data.

Embodiment 26. The method of Embodiment 1, wherein determining, based on the prediction scores, whether the GAN has been trained comprises comparing one or more prediction scores with a threshold value.

Embodiment 27. A method for training a generative adversarial network (GAN) comprising: generating increasingly accurate positive simulated polypeptide-MHC-I interaction data by the GAN generator until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive; presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC interaction data -I as positive or negative; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN to generate predictive scores; determining, based on the prediction estimates, that the GAN is not trained; repeating steps a-c until a determination is made based on prediction estimates that the GAN has been trained; and derivation of GANs and CNNs.

Embodiment 28. The method of Embodiment 27, wherein generating increasingly fine positive polypeptide-MHC-I interaction model data by the GAN generator until the GAN discriminator classifies the positive polypeptide-MHC-I interaction model as positive, comprises: generating, by means of the GAN generator, in accordance with the GAN parameter set, a first simulated dataset containing simulated positive polypeptide-MHC-I interactions for the MHC allele; combining the first simulated data set with positive real polypeptide-MHC-I interactions for the MHC allele and negative real polypeptide-MHC-I interactions for the MHC allele to create a GAN training data set; determining, with a discriminator according to a decision boundary, whether a positive polypeptide-MHC-I interaction for an MHC allele in the GAN training dataset is a simulated positive, a real positive, or a real negative; an adjustment based on the accuracy of the discriminator's determination of one or more of the GAN parameter set or decision boundary; and repeating steps g-j until the first stop criterion is met.

Embodiment 29. The method of Embodiment 28, wherein presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, provides for: generating, by means of a GAN generator in accordance with the GAN parameter set, a second simulated dataset containing simulated positive polypeptide-MHC-I interactions for the MHC allele; combining the second simulated dataset, the known positive polypeptide-MHC-I interactions for the MHC allele, and the known negative polypeptide-MHC-I interactions for the MHC allele to create a CNN training dataset; representing a CNN training dataset into a convolutional neural network (CNN); classifying by the CNN according to the polypeptide-MHC-I interaction CNN parameter set for the MHC allele in the CNN training data set as positive or negative; an adjustment based on the classification accuracy by the CNN, one or more of the set of CNN parameters; and repeating steps n-p until the second stopping criterion is met.

Embodiment 30. The method of Embodiment 29, wherein representing positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in a CNN to generate predictive scores comprises: classifying by the CNN according to a set of polypeptide interaction CNN parameters -MHC-I for MHC allele as positive or negative.

Embodiment 31. The method of Embodiment 30, wherein determining, based on the prediction scores, that the GAN has been trained comprises determining the classification accuracy by the CNN, wherein when (if) the classification accuracy satisfies the third stopping criterion, deriving the GAN and the CNN.

Embodiment 32. The method of Embodiment 31, wherein determining, based on the prediction estimates that the GAN is trained, comprises determining the classification accuracy by the CNN, wherein when (if) the classification accuracy does not satisfy the third stopping criterion, returning to step a.

Embodiment 33. The method of Embodiment 28, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 34. The method of Embodiment 33, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 35. The method of Embodiment 33, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 36. The method of Embodiment 35, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 37. The method of Embodiment 27, further comprising: presenting a data set to a CNN, where the data set contains a plurality of candidate polypeptide-MHC-I interactions; classifying by CNN each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified as a positive polypeptide-MHC-I interaction.

Embodiment 38 A polypeptide obtained by the method of Embodiment 37.

Embodiment 39. The method of Embodiment 37 wherein the polypeptide is a tumor specific antigen.

Embodiment 40. The method of Embodiment 37, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele.

Embodiment 41. The method of Embodiment 27, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with the selected allele.

Embodiment 42. The method of Embodiment 41, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 43. The method of Embodiment 27, wherein generating increasingly accurate positive simulated polypeptide-MHC-I interaction data by the GAN generator until the GAN discriminator classifies the positive simulated polypeptide-MHC-I interaction data as positive, involves evaluating gradient descent expressions for the GAN generator.

Embodiment 44. The method of Embodiment 27, wherein generating increasingly fine positive polypeptide-MHC-I interaction model data by the GAN generator until the GAN discriminator classifies the positive polypeptide-MHC-I interaction model as positive, comprises: iteratively performing (eg, optimizing) the GAN discriminator to increase the ability to assign a high probability to positive real polypeptide-MHC-I interaction data, a low probability to positive simulated polypeptide-MHC-I interaction, and a low probability to negative real polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the likelihood that positive simulated polypeptide-MHC-I interaction data will score high.

Embodiment 45. The method of Embodiment 27, wherein presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, provides for: performing a convolution procedure; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 46. The method of Embodiment 27, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 47. The method of Embodiment 28, wherein the first stopping criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 48. The method of Embodiment 27, wherein the second stopping criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 49. The method of embodiment 31 or 32, wherein the third stopping criterion includes an estimate of the area under the curve (AUC) function.

Embodiment 50. The method of Embodiment 27, wherein the predictive score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data.

Embodiment 51. The method of Embodiment 27, wherein determining, based on the prediction scores, whether the GAN has been trained comprises comparing one or more prediction scores with a threshold value.

Embodiment 52. A method for training a generative adversarial network (GAN), comprising: generating, by means of a GAN generator, in accordance with a set of GAN parameters, a first simulated dataset containing simulated positive polypeptide-MHC-I interactions for an MHC allele; combining the first simulated dataset with positive real polypeptide-MHC-I interactions and negative real polypeptide-MHC-I interactions for the MHC allele; determining with a discriminator according to a decision boundary whether a positive polypeptide-MHC-I interaction for an MHC allele in the GAN training dataset is positive or negative; an adjustment based on the accuracy of the discriminator's determination of one or more of the GAN parameter set or decision boundary; repeating steps a-d until the first stop criterion is met; generating by the GAN generator in accordance with the GAN parameter set a second simulated dataset containing simulated positive interactions of the polypeptide-MHC-I for the MHC allele; combining a second simulated dataset, positive real polypeptide-MHC-I interactions, and negative real polypeptide-MHC-I interactions to create a CNN training dataset; representing a CNN training dataset into a convolutional neural network (CNN); classifying by the CNN according to the polypeptide-MHC-I interaction CNN parameter set for the MHC allele in the CNN training data set as positive or negative; adjusting based on classification accuracy by the polypeptide-MHC-I interaction CNN for the MHC allele in the CNN training data set, one or more of the CNN parameter set; repeating steps h-j until the second stop criterion is met; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN; classifying the CNN according to the polypeptide-MHC-I interaction CNN parameter set for the MHC allele as positive or negative; and determining the classification accuracy by means of the polypeptide-MHC-I interaction CNN for the MHC allele, wherein when (if) the classification accuracy satisfies the third stopping criterion, outputting the GAN and CNN, wherein when (if) the classification accuracy does not satisfy the third stopping criterion, return to step a.

Embodiment 53. The method of Embodiment 52, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 54. The method of Embodiment 52 wherein the MHC allele is an HLA allele.

Embodiment 55. The method of Embodiment 54, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 56. The method of Embodiment 54 wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 57. The method of Embodiment 54 wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 58. The method of Embodiment 52, further comprising: presenting a data set to a CNN, where the data set contains a plurality of candidate polypeptide-MHC-I interactions; classifying by CNN each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified as a positive polypeptide-MHC-I interaction.

Embodiment 59 A polypeptide obtained by the method of Embodiment 58.

Embodiment 60. The method of Embodiment 58, wherein the polypeptide is a tumor-specific antigen.

Embodiment 61. The method of Embodiment 58, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected human leukocyte antigen (HLA) allele.

Embodiment 62. The method of Embodiment 52, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with the selected allele.

Embodiment 63. The method of Embodiment 62, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 64. The method of Embodiment 52, wherein repeating steps a-d until the first stopping criterion is met, evaluates the gradient descent expression for the GAN generator.

Embodiment 65. The method of Embodiment 52 wherein repeating steps a-d until the first stop criterion is met comprises: iteratively executing (e.g., optimizing) the GAN discriminator to increase the ability to assign a high probability to positive real polypeptide-MHC interaction data -I, low probability of positive simulated polypeptide-MHC-I interaction data and low probability of negative real polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the likelihood that positive simulated polypeptide-MHC-I interaction data will score high.

Embodiment 66. The method of Embodiment 52, wherein the CNN's representation of data for training in the CNN includes: performing a convolution procedure, performing a non-linearity procedure (e.g., ReLu), performing a pooling or downsampling procedure, and performing a classification procedure (e.g., fully connected layer).

Embodiment 67. The method of Embodiment 52, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 68. The method of Embodiment 52, wherein the first stopping criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 69. The method of Embodiment 52, wherein the second stopping criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 70. The method of Embodiment 52, wherein the third stopping criterion includes an estimate of an area under the curve (AUC) function.

Embodiment 71. A method comprising: training a convolutional neural network (CNN) according to the method of Embodiment 1; representing a dataset in a CNN, where the dataset contains a plurality of candidate polypeptide-MHC-I interactions; classifying by CNN each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide associated with a candidate polypeptide-MHC-I interaction classified as a positive polypeptide-MHC-I interaction.

Embodiment 72. The method of Embodiment 71, wherein the CNN is trained based on one or more GAN parameters including one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 73. The method of Embodiment 72, wherein the allele type is the HLA allele type.

Embodiment 74. The method of Embodiment 73, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 75. The method of Embodiment 73 wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 76. The method of Embodiment 73, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 77 A polypeptide obtained by the method of Embodiment 71.

Embodiment 78. The method of Embodiment 71 wherein the polypeptide is a tumor specific antigen.

Embodiment 79. The method of Embodiment 71, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected human leukocyte antigen (HLA) allele.

Embodiment 80. The method of Embodiment 71, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with the selected allele.

Embodiment 81. The method of Embodiment 80, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 82. The method of Embodiment 71, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 83. A device for training a generative adversarial network (GAN), comprising: one or more processors; and a memory that stores processor-executable instructions that, when executed by one or more processors, cause the device to: generate increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide interaction data. -MHC-I as positive; presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC interaction data -I as positive or negative; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN to generate predictive scores; determining, based on prediction estimates, that the GAN is trained; and derivation of GAN and CNN.

Embodiment 84. The apparatus of Embodiment 83, wherein processor-executable instructions that, when executed by one or more processors, cause the device to generate increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated interaction data polypeptide-MHC-I as positive, further provide processor-executable instructions that, when executed by one or more processors, cause the device to do the following: generate, in accordance with the GAN parameter set, a first simulated data set containing simulated positive polypeptide-MHC-I interactions for the MHC allele ; combining the first simulated dataset of positive real interactions polypeptide-MHC-I for the MHC allele and negative real interactions polypeptide-MHC-I for the MHC allele to create a GAN training dataset; obtaining information from the discriminator, the discriminator being configured to determine, according to a decision boundary, whether a positive interaction of the polypeptide-MHC-I for the MHC allele in the GAN training data set is positive or negative; an adjustment based on the accuracy of the information from the discriminator, one or more of the GAN parameter set, or the decision boundary; and repeating steps a-d until the first stop criterion is met.

Embodiment 85. The apparatus of Embodiment 84, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide interaction data -MHC-I to a Convolutional Neural Network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, cause the device to do the following: in accordance with the GAN parameter set of the second simulated dataset containing the simulated positive interactions of the polypeptide-MHC-I for the MHC allele; combining a second simulated dataset, positive real-life polypeptide-MHC-I interactions for the MHC allele, and negative real-world polypeptide-MHC-I interaction data for the MHC allele to create a CNN training dataset; representing a CNN training dataset into a convolutional neural network (CNN); obtaining training information from the CNN, the CNN being configured to determine the training information by classifying according to the polypeptide-MHC-I interaction parameter set of the CNN for the MHC allele in the training data set of the CNN as positive or negative; adjusting, based on the accuracy of the training information, one or more of the set of CNN parameters; and repeating steps h-j until the second stop criterion is met.

Embodiment 86. The apparatus of Embodiment 85, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in a CNN to generate predictive estimates , further provide processor-executable instructions that, when executed by one or more processors, cause the device to classify, according to a set of polypeptide-MHC-I interaction parameters, for an MHC allele as positive or negative.

Embodiment 87. The apparatus of Embodiment 86, wherein processor-executable instructions that, when executed by one or more processors, cause the device to determine, based on prediction estimates, whether the GAN has been trained, further provides processor-executable instructions that, when executed by one or more processors, determine a polypeptide-MHC-I interaction classification accuracy device for the MHC allele as positive or negative, and when (if) the classification accuracy satisfies the third stopping criterion, inferring GANs and CNNs.

Embodiment 88. The apparatus of Embodiment 86, wherein processor-executable instructions that, when executed by one or more processors, cause the device to determine, based on prediction estimates, whether the GAN has been trained, further provides processor-executable instructions that, when executed by one or more processors, determine a polypeptide-MHC-I interaction classification accuracy device for the MHC allele as positive or negative, and when (if) the classification accuracy does not satisfy the third stopping criterion, return to step a.

Embodiment 89. The apparatus of embodiment 84, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 90. The device of Embodiment 89, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 91. The device of Embodiment 89, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 92. The device of Embodiment 89, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 93. The apparatus of Embodiment 83, wherein the processor-executable instructions, when executed by one or more processors, further cause the apparatus to perform the following: present a dataset in a CNN, wherein the dataset contains a plurality of candidate polypeptide-MHC-I interactions, wherein the CNN further configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction that CNN classifies as a positive polypeptide-MHC-I interaction.

Embodiment 94 A polypeptide produced using the device of Embodiment 93.

Embodiment 95. The device of Embodiment 93, wherein the polypeptide is a tumor-specific antigen.

Embodiment 96. The device of Embodiment 93, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected human leukocyte antigen (HLA) allele.

Embodiment 97. The device of embodiment 83, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 98. The device of Embodiment 97 wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 99. The apparatus of Embodiment 83, wherein processor-executable instructions that, when executed by one or more processors, cause the device to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated interaction data polypeptide-MHC-I as positive further provide processor-executable instructions that, when executed by one or more processors, cause the device to evaluate the gradient descent expression for the GAN generator.

Embodiment 100. The apparatus of embodiment 83, wherein processor-executable instructions that, when executed by one or more processors, cause the device to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated interaction data polypeptide-MHC-I as positive further provide processor-executable instructions that, when executed by one or more processors, cause the device to perform the following: iteratively executes (e.g., optimizes) the GAN discriminator to increase the ability to assign a high probability to positive real-life polypeptide-MHC-I interaction data , low probability of positive simulated polypeptide-MHC-I interaction data and low probability of negative simulated polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the possibility that positive simulated polypeptide-MHC-I interaction data will be highly valued.

Embodiment 101. The apparatus of Embodiment 83, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive simulated polypeptide-MHC-I interaction data, positive real-world polypeptide-MHC-I interaction data, and negative real-world polypeptide interaction data -MHC-I to a Convolutional Neural Network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, cause the device to do the following: convolution procedures;

performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure, and performing a classification procedure (fully connected layer).

Embodiment 102. The apparatus of Embodiment 83, wherein the GAN includes a Deep Convolutional GAN (DCGAN).

Embodiment 103. The apparatus of embodiment 84, wherein the first stop criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 104. The apparatus of embodiment 85, wherein the second stop criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 105. The apparatus of embodiment 87 or 88, wherein the third stopping criterion includes an estimate of the area under the curve (AUC) function.

Embodiment 106. The apparatus of embodiment 83, wherein the prediction score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data.

Embodiment 107. The apparatus of Embodiment 83, wherein processor-executable instructions that, when executed by one or more processors, cause the device to determine, based on prediction estimates, whether the GAN has been trained, further provides processor-executable instructions that, when executed by one or more processors, provide a comparison a device of one or more threshold prediction estimates.

Embodiment 108. A device for training a generative adversarial network (GAN), comprising:

one or more processors and

a storage device that stores processor-executable instructions that, when executed by one or more processors, cause the device to do the following: generate increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data MHC-I as positive; presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC interaction data -I as positive or negative; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN to generate predictive scores; determining, based on the prediction estimates, that the GAN is not trained; repeating steps a-c until a determination is made based on prediction estimates that the GAN has been trained; and derivation of GAN and CNN.

Embodiment 109. The apparatus of embodiment 108, wherein processor-executable instructions that, when executed by one or more processors, cause the device to generate increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated interaction data polypeptide-MHC-I as positive, further provide processor-executable instructions that, when executed by one or more processors, cause the device to do the following: generate, in accordance with the GAN parameter set, a first simulated data set containing simulated positive polypeptide-MHC-I interactions for the MHC allele ; combining the first simulated data set with positive real polypeptide-MHC-I interactions for the MHC allele and positive real polypeptide-MHC-I interactions for the MHC allele to create a GAN training data set; obtaining information from the discriminator, the discriminator being configured to determine whether a positive interaction of the polypeptide-MHC-I for the MHC allele in the GAN training data set is positive or negative; an adjustment based on the accuracy of the information from the discriminator, one or more of the GAN parameter set, or the decision boundary; and repeating steps i-j until the first stop criterion is met.

Embodiment 110. The apparatus of embodiment 109, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive simulated polypeptide-MHC-I interaction data, positive real-world polypeptide-MHC-I interaction data, and negative real-world polypeptide interaction data -MHC-I to a Convolutional Neural Network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, cause the device to do the following: in accordance with the GAN parameter set of the second simulated dataset containing the simulated positive interactions of the polypeptide-MHC-I for the MHC allele; combining the second simulated dataset, positive real polypeptide-MHC-I interactions, and negative real polypeptide-MHC-I interaction data to create a CNN training dataset; representing a dataset for training a CNN into a convolutional neural network (CNN); obtaining information from the CNN, wherein the CNN is configured to determine the information by classifying according to a set of polypeptide-MHC-I interaction parameters of the CNN for the MHC allele in the CNN training dataset as positive or negative; an adjustment based on the accuracy of the training information from the CNN, one or more of the set of CNN parameters; and repeating steps n-p until the second stopping criterion is met.

Embodiment 111. The apparatus of embodiment 110, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in a CNN to generate predictive estimates , further provide processor-executable instructions that, when executed by one or more processors, cause the device to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN, wherein the CNN is further configured to be classified according to a set polypeptide-MHC-I interaction CNN parameters for the MHC allele as positive or negative.

Embodiment 112. The apparatus of embodiment 111, wherein processor-executable instructions that, when executed by one or more processors, cause the device to determine, based on prediction estimates, whether the GAN has been trained, further provides processor-executable instructions that, when executed by one or more processors, cause the device to execute the device of the following: determining the accuracy of classification by means of CNN; determining that the classification accuracy satisfies the third stopping criterion; and in response to determining that the classification accuracy satisfies the third stopping criterion, deriving the GAN and CNN.

Embodiment 113. The apparatus of embodiment 112, wherein processor-executable instructions that, when executed by one or more processors, cause the device to determine, based on prediction estimates, whether the GAN has been trained, further provides processor-executable instructions that, when executed by one or more processors, cause the device to execute the device of the following: determining the accuracy of classification by means of CNN; determining that the classification accuracy does not satisfy the third stopping criterion; and in response to determining that the classification accuracy does not satisfy the third stopping criterion, returning to step a.

Embodiment 114. The apparatus of embodiment 109, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 115. The device of embodiment 109, wherein the MHC allele is an HLA allele.

Embodiment 116. The device of embodiment 115, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 117. The device of embodiment 115, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 118. The device of embodiment 115, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 119. The apparatus of embodiment 108, wherein the processor-executable instructions, when executed by one or more processors, further cause the apparatus to: present a dataset in a CNN, wherein the dataset contains a plurality of candidate polypeptide-MHC-I interactions, wherein the CNN further configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified by the CNN as a positive polypeptide-MHC-I interaction.

Embodiment 120. Polypeptide produced using the device of Embodiment 119.

Embodiment 121. The device of Embodiment 119, wherein the polypeptide is a tumor-specific antigen.

Embodiment 122. The device of embodiment 119, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected human leukocyte antigen (HLA) allele.

Embodiment 123. The device of embodiment 108, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 124. The device of embodiment 123 wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 125. The apparatus of embodiment 108, wherein processor-executable instructions that, when executed by one or more processors, cause the device to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated interaction data polypeptide-MHC-I as positive further provide processor-executable instructions that, when executed by one or more processors, cause the device to evaluate the gradient descent expression for the GAN generator.

Embodiment 126. The apparatus of embodiment 108, wherein processor-executable instructions that, when executed by one or more processors, cause the device to generate increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated interaction data polypeptide-MHC-I as positive further provide processor-executable instructions that, when executed by one or more processors, cause the device to perform the following: , low probability of positive simulated polypeptide-MHC-I interaction data and low probability of negative simulated polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the possibility that positive simulated polypeptide-MHC-I interaction data will be highly valued.

Embodiment 127. The apparatus of embodiment 108, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide interaction data -MHC-I to a Convolutional Neural Network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, cause the device to do the following: convolution procedures; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 128. The apparatus of embodiment 108, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 129. The apparatus of embodiment 109, wherein the first stop criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 130. The apparatus of embodiment 108, wherein the second stop criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 131. The apparatus of embodiment 112 or 113, wherein the third stopping criterion includes an estimate of an area under the curve (AUC) function.

Embodiment 132. The apparatus of embodiment 108, wherein the predictive score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data.

Embodiment 133. The apparatus of embodiment 108, wherein processor-executable instructions that, when executed by one or more processors, cause the device to determine, based on prediction estimates, whether the GAN has been trained, further provides processor-executable instructions that, when executed by one or more processors, provide a comparison a device of one or more threshold prediction estimates.

Embodiment 134. A device for training a generative adversarial network (GAN), comprising: one or more processors; and a memory that stores processor-executable instructions that, when executed by one or more processors, cause the device to: generate, in accordance with the GAN parameter set, a first simulated data set containing simulated positive interactions of the polypeptide-MHC-I for the MHC allele; combining the first simulated dataset with positive real polypeptide-MHC-I interactions for the MHC allele and negative real polypeptide-MHC-I interactions for the MHC allele to create a GAN training dataset; obtaining information from the discriminator, the discriminator being configured to determine, according to a decision boundary, whether a positive interaction of the polypeptide-MHC-I for the MHC allele in the GAN training data set is positive or negative; an adjustment based on the accuracy of the information from the discriminator, one or more of the GAN parameter set, or the decision boundary;

repeating steps a-d until the first stop criterion is met; using the GAN generator, according to the GAN parameter set, generating a second simulated dataset containing simulated positive polypeptide-MHC-I interactions for the MHC allele; combining the second simulated data set, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data for the MHC allele to create a CNN training data set; representing a dataset for training a CNN into a convolutional neural network (CNN); obtaining training information from the CNN, wherein the CNN is configured to determine the training information by classifying according to the polypeptide-MHC-I interaction parameter set of the CNN for the MHC allele in the training data set of the CNN as positive or negative; adjusting, based on the accuracy of the training information, one or more of the set of CNN parameters; repeating steps h-j until the second stop criterion is met; presentation in the CNN of positive real interactions polypeptide-MHC-I for the MHC allele and negative real interactions polypeptide-MHC-I for the MHC allele; obtaining information for training from the CNN, where the CNN is configured to determine the training information by classifying according to a set of CNN interaction parameters of the polypeptide-MHC-I for the MHC allele as positive or negative; and determining the accuracy of the training information, wherein when (if) the accuracy of the training information satisfies the third stopping criterion, deriving the GAN and CNN, and when (if) the accuracy of the training information does not satisfy the third stopping criterion, returning to step a.

Embodiment 135. The apparatus of embodiment 134, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 136. The device of embodiment 134, wherein the MHC allele is an HLA allele.

Embodiment 137. The device of embodiment 136 wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 138. The device of embodiment 136, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 139. The device of Embodiment 136, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 140. The apparatus of embodiment 134, wherein the processor-executable instructions, when executed by one or more processors, further cause the apparatus to: present a dataset in a CNN, wherein the dataset contains a plurality of candidate polypeptide-MHC-I interactions, wherein the CNN further configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified by the CNN as a positive polypeptide-MHC-I interaction.

Embodiment 141 A polypeptide produced using the device of Embodiment 140.

Embodiment 142. The device of embodiment 140, wherein the polypeptide is a tumor-specific antigen.

Embodiment 143. The device of embodiment 140, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the MHC allele.

Embodiment 144. The apparatus of embodiment 134, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 145. The device of embodiment 144 wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 146. The apparatus of embodiment 134, wherein processor-executable instructions that, when executed by one or more processors, cause the device to repeat steps a-d until the first stop criterion is met, further provide processor-executable instructions that, when executed by one or more more processors provide device evaluation of the gradient descent expression for the GAN generator.

Embodiment 147. The apparatus of embodiment 134, wherein processor-executable instructions that, when executed by one or more processors, cause the device to repeat steps a-d until the first stop criterion is met, further provide processor-executable instructions that, when executed by one or more more processors cause the device to: iteratively execute (e.g., optimize) the GAN discriminator to increase the ability to assign a high probability of positive real polypeptide-MHC-I interaction data, a low probability of positive simulated polypeptide-MHC-I interaction data, and a low probability of negative simulated interaction data polypeptide-MHC-I; and iteratively performing (eg, optimizing) the GAN generator to increase the possibility that positive simulated polypeptide-MHC-I interaction data will be highly valued.

Embodiment 148. The apparatus of embodiment 134, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present data to the CNN for training the CNN, further provide processor-executable instructions that, when executed by one or more processors, cause the device to perform the following: performing the convolution procedure; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 149. The apparatus of embodiment 134, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 150. The apparatus of embodiment 134, wherein the first stop criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 151. The apparatus of embodiment 134, wherein the second stop criterion includes an estimate of a Mean Square Error (MSE) function.

Embodiment 152. The apparatus of embodiment 134, wherein the third stopping criterion includes an estimate of an area under the curve (AUC) function.

Embodiment 153. An apparatus comprising: one or more processors; and a memory that stores processor-executable instructions that, when executed by one or more processors, cause the device to: train a convolutional neural network (CNN) by the same means as the device of embodiment 83; presenting the dataset to a CNN, where the dataset comprises a plurality of candidate polypeptide-MHC-I interactions, wherein the CNN is configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide associated with a candidate polypeptide-MHC-I interaction classified by the CNN as a positive polypeptide-MHC-I interaction.

Embodiment 154. The apparatus of embodiment 153, wherein the CNN is trained based on one or more GAN parameters including one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 155. The device of embodiment 154, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 156. The device of embodiment 154, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 157. The device of embodiment 155, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 158 A polypeptide produced using the device of Embodiment 153.

Embodiment 159. The device of Embodiment 153, wherein the polypeptide is a tumor-specific antigen.

Embodiment 160. The device of Embodiment 153, wherein the polypeptide comprises an amino acid sequence that specifically binds to an MHC-I protein encoded by the selected MHC allele.

Embodiment 161. The device of embodiment 153, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 162. The device of embodiment 161 wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 163. The apparatus of embodiment 153, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 164. A non-volatile computer-readable medium for training a generative adversarial network (GAN), wherein the non-volatile computer-readable medium stores processor-executable instructions that, when executed by one or more processors, cause the one or more processors to do the following: generate increasingly accurate positive simulated interaction data polypeptide -MHC-I until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive; presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC interaction data -I as positive or negative; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN to generate predictive scores; determining, based on prediction estimates, that the GAN is trained; and derivation of GAN and CNN.

Embodiment 165. The non-volatile computer-readable medium of embodiment 164, wherein processor-executable instructions that, when executed by one or more processors, cause the one or more processors to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifying positive simulated polypeptide-MHC-I interaction data as positive, further causing one or more processors to: generate, in accordance with the GAN parameter set, a first simulated data set containing simulated positive polypeptide-MHC-I interactions for the MHC allele; combining the first simulated data set with positive real polypeptide-MHC-I interactions for the MHC allele and negative real polypeptide-MHC-I interactions for the MHC allele to create a GAN training data set; obtaining information from the discriminator, the discriminator being configured to determine, according to a decision boundary, whether a positive interaction of the polypeptide-MHC-I for the MHC allele in the GAN training data set is positive or negative; an adjustment based on the accuracy of the information from the discriminator, one or more of the GAN parameter set, or the decision boundary; and repeating steps a-d until the first stop criterion is met.

Embodiment 166. The nonvolatile computer-readable medium of embodiment 165, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to present positive simulated polypeptide-MHC-I interaction data, positive real-world polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, provide performing by one or more processors the following: generating, in accordance with the GAN parameter set, a second simulated data set containing simulated positive interactions of the polypeptide-MHC-I; combining a second simulated dataset, positive real-life polypeptide-MHC-I interactions for the MHC allele, and negative real-life polypeptide-MHC-I interaction data for the MHC allele to create a CNN training dataset; representing a dataset for training a CNN into a convolutional neural network (CNN); obtaining training information from the CNN, the CNN being configured to determine the training information by classifying according to the polypeptide-MHC-I interaction parameter set of the CNN for the MHC allele in the training data set of the CNN as positive or negative; adjusting, based on the accuracy of the training information, one or more of the set of CNN parameters; and repeating steps h-j until the second stopping criterion is met.

Embodiment 167. The nonvolatile computer-readable medium of embodiment 166, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in The CNNs for generating prediction scores further provide processor-executable instructions that, when executed by one or more processors, cause the one or more processors to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN, where the CNN is further configured to classify, according to a set of CNN interaction parameters, the polypeptide-MHC-I for the MHC allele as positive or negative.

Embodiment 168. The nonvolatile computer-readable medium of embodiment 167, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to determine, based on prediction estimates, whether the GAN has been trained, further provide processor-executable instructions that, when executed by one or more or more processors cause one or more processors to determine the classification accuracy of the polypeptide-MHC-I interaction for the MHC allele as positive or negative and, when (if) the classification accuracy satisfies the third stop criterion, inferring GANs and CNNs.

Embodiment 169. The nonvolatile computer-readable medium of embodiment 167, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to determine, based on prediction estimates, whether the GAN has been trained, further provide processor-executable instructions that, when executed by one or more or more processors cause the one or more processors to determine the classification accuracy of the polypeptide-MHC-I interaction for the MHC allele as positive or negative and, when (if) the classification accuracy does not satisfy the third stop criterion, return to step a.

Embodiment 170. The nonvolatile computer-readable medium of embodiment 165, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 171. The non-volatile computer-readable medium of embodiment 165, wherein the MHC allele is an HLA allele.

Embodiment 172. The nonvolatile computer-readable medium of Embodiment 171, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 173. The nonvolatile computer-readable medium of Embodiment 171, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 174. The nonvolatile computer-readable medium of Embodiment 171, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 175. The nonvolatile computer-readable medium of embodiment 164, wherein the processor-executable instructions, when executed by one or more processors, further cause the one or more processors to perform the following: Present a dataset in a CNN, wherein the dataset contains a plurality of candidate polypeptide-MHC-I interactions wherein the CNN is further configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction that CNN classifies as a positive polypeptide-MHC-I interaction.

Embodiment 176. The polypeptide obtained using the non-volatile computer-readable medium of Embodiment 175.

Embodiment 177. The nonvolatile computer-readable medium of Embodiment 175, wherein the polypeptide is a tumor-specific antigen.

Embodiment 178. The nonvolatile computer-readable medium of embodiment 175, wherein the polypeptide contains an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele.

Embodiment 179. The nonvolatile computer-readable medium of embodiment 164, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 180. The nonvolatile computer-readable medium of Embodiment 179, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 181. The non-volatile computer-readable medium of embodiment 164, wherein processor-executable instructions that, when executed by one or more processors, cause the one or more processors to generate increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifying positive simulated polypeptide-MHC-I interaction data as positive, further providing processor-executable instructions that, when executed by one or more processors, cause one or more processors to evaluate a gradient descent expression for the GAN generator.

Embodiment 182. The nonvolatile computer-readable medium of embodiment 164, wherein processor-executable instructions that, when executed by one or more processors, cause the one or more processors to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive, further provides for processor-executable instructions that, when executed by one or more processors, causes one or more processors to perform the following: real polypeptide-MHC-I interaction data and low probability positive simulated polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the possibility that positive simulated polypeptide-MHC-I interaction data will be highly valued.

Embodiment 183. The nonvolatile computer-readable medium of embodiment 164, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to present positive simulated polypeptide-MHC-I interaction data, positive real-world polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as real positive or negative, further provide processor-executable instructions that, when executed by one or more processors cause one or more processors to perform the following: perform a convolution procedure; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 184. The non-volatile computer-readable medium of embodiment 164, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 185. The non-volatile computer-readable medium of embodiment 165, wherein the first stop criterion includes an estimate of a mean square error (MSE) function.

Embodiment 186. The non-volatile computer-readable medium of embodiment 166, wherein the second stop criterion includes an estimate of a mean square error (MSE) function.

Embodiment 187. The nonvolatile computer-readable medium of embodiment 168 or 169, wherein the third stop criterion includes an area under the curve (AUC) function estimate.

Embodiment 188. The nonvolatile computer-readable medium of embodiment 164, wherein the prediction score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data.

Embodiment 189. The nonvolatile computer-readable medium of embodiment 164, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to determine, based on prediction estimates, whether the GAN has been trained, further provide processor-executable instructions that, when executed by one or more or more processors cause one or more processors to compare one or more prediction estimates with a threshold value.

Embodiment 190. A non-volatile computer-readable medium for training a generative adversarial network (GAN), wherein the non-volatile computer-readable medium stores processor-executable instructions that, when executed by one or more processors, cause the one or more processors to do the following: generate increasingly accurate positive simulated interaction data polypeptide -MHC-I until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive; presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC interaction data -I as positive or negative; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN to generate predictive scores; determining, based on the prediction estimates, that the GAN is not trained; repeating steps a-c until a determination is made based on prediction estimates that the GAN has been trained; and derivation of GAN and CNN.

Embodiment 191. The non-volatile computer-readable medium of embodiment 190, wherein processor-executable instructions that, when executed by one or more processors, cause the one or more processors to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive, further provides processor-executable instructions that, when executed by one or more processors, cause one or more processors to: generate, in accordance with the GAN parameter set, a first simulated data set containing simulated positive interactions polypeptide-MHC-I for the MHC allele; combining the first simulated data set with positive real polypeptide-MHC-I interactions for the MHC allele and positive real polypeptide-MHC-I interactions for the MHC allele to create a GAN training data set; obtaining information from the discriminator, the discriminator being configured to determine whether a positive interaction of the polypeptide-MHC-I for the MHC allele in the GAN training data set is positive or negative; an adjustment based on the accuracy of the information from the discriminator, one or more of the GAN parameter set, or the decision boundary; and repeating steps g-j until the first stopping criterion is met.

Embodiment 192. The nonvolatile computer-readable medium of embodiment 191, wherein processor-executable instructions that, when executed by one or more processors, cause the device to present positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real data polypeptide-MHC-I interactions to a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, cause one or more more processors of the following: generating, in accordance with the GAN parameter set, a second simulated data set containing simulated positive polypeptide-MHC-I interactions; combining a second simulated dataset, positive real-life polypeptide-MHC-I interactions for the MHC allele, and negative real-world polypeptide-MHC-I interaction data for the MHC allele to create a CNN training dataset; representing a dataset for training a CNN into a convolutional neural network (CNN); obtaining information from the CNN, wherein the CNN is configured to determine the information by classifying according to a set of polypeptide-MHC-I interaction parameters of the CNN for the MHC allele in the CNN training dataset as positive or negative; an adjustment based on the accuracy of the training information from the CNN, one or more of the set of CNN parameters; and repeating steps l-p until the second stop criterion is met.

Embodiment 193. The nonvolatile computer-readable medium of embodiment 192, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in The CNNs for generating prediction scores further provide processor-executable instructions that, when executed by one or more processors, cause the one or more processors to present positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN, where the CNN is further configured to classify, according to a set of CNN interaction parameters, the polypeptide-MHC-I interaction for the MHC allele as positive or negative.

Embodiment 194. The nonvolatile computer-readable medium of embodiment 193, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to determine, based on prediction estimates, whether the GAN has been trained, further provide processor-executable instructions that, when executed by one or more or more processors cause one or more processors to perform the following: determining the accuracy of the classification by the CNN; determining that the classification accuracy satisfies the third stopping criterion; and in response to determining that the classification accuracy satisfies the third stopping criterion, deriving the GAN and CNN.

Embodiment 195. The nonvolatile computer-readable medium of embodiment 194, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to determine, based on prediction estimates, whether the GAN has been trained, further provide processor-executable instructions that, when executed by one or more or more processors cause one or more processors to perform the following: determining the accuracy of the classification by the CNN; determining that the classification accuracy does not satisfy the third stopping criterion; and in response to determining that the classification accuracy does not satisfy the third stopping criterion, returning to step a.

Embodiment 196. The nonvolatile computer-readable medium of embodiment 191, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 197. The non-volatile computer-readable medium of Embodiment 191, wherein the MHC allele is an HLA allele.

Embodiment 198. The nonvolatile computer-readable medium of embodiment 197, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 199. The nonvolatile computer-readable medium of Embodiment 197, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 200. The non-volatile computer-readable medium of Embodiment 197, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 201. The nonvolatile computer-readable medium of embodiment 190, wherein the processor-executable instructions, when executed by one or more processors, further cause the one or more processors to: present a dataset to a CNN, wherein the dataset contains a plurality of candidate polypeptide-MHC-I interactions wherein the CNN is further configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified by the CNN as a positive polypeptide-MHC-I interaction.

Embodiment 202. A polypeptide obtained using the non-volatile computer-readable medium of Embodiment 201.

Embodiment 203. The non-volatile computer-readable medium of Embodiment 201, wherein the polypeptide is a tumor-specific antigen.

Embodiment 204. The nonvolatile computer-readable medium of embodiment 201, wherein the polypeptide contains an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele.

Embodiment 205. The nonvolatile computer-readable medium of embodiment 190, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 206. The nonvolatile computer-readable medium of embodiment 205, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 207. The non-volatile computer-readable medium of embodiment 190, wherein processor-executable instructions that, when executed by one or more processors, cause the one or more processors to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifying positive simulated polypeptide-MHC-I interaction data as positive, further providing processor-executable instructions that, when executed by one or more processors, cause one or more processors to evaluate a gradient descent expression for the GAN generator.

Embodiment 208. The nonvolatile computer-readable medium of embodiment 190, wherein processor-executable instructions that, when executed by one or more processors, cause the one or more processors to generate progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as positive, further provides for processor-executable instructions that, when executed by one or more processors, causes one or more processors to perform the following: real polypeptide-MHC-I interaction data, low probability of positive simulated polypeptide-MHC-I interaction data, and low probability of negative simulated interaction data polypeptide-MHC-I; and iteratively performing (eg, optimizing) the GAN generator to increase the possibility that positive simulated polypeptide-MHC-I interaction data will be highly valued.

Embodiment 209. The nonvolatile computer-readable medium of embodiment 190, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to present positive simulated polypeptide-MHC-I interaction data, positive real-world polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data into a convolutional neural network (CNN) until the CNN classifies the polypeptide-MHC-I interaction data as positive or negative, further provide processor-executable instructions that, when executed by one or more processors, provide performing by one or more processors the following: performing a convolution procedure; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 210. The non-volatile computer-readable medium of embodiment 190, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 211. The nonvolatile computer-readable medium of embodiment 191, wherein the first stop criterion includes an estimate of a mean square error (MSE) function.

Embodiment 212. The nonvolatile computer-readable medium of embodiment 190, wherein the second stop criterion includes an estimate of a mean square error (MSE) function.

Embodiment 213. The nonvolatile computer-readable medium of embodiment 194 or 195, wherein the third stopping criterion includes an area under the curve (AUC) function estimate.

Embodiment 214. The nonvolatile computer-readable medium of embodiment 190, wherein the prediction score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data.

Embodiment 215. The nonvolatile computer-readable medium of embodiment 190, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to determine, based on prediction estimates, whether the GAN has been trained, further provide processor-executable instructions that, when executed by one or more or more processors cause one or more processors to compare one or more prediction estimates with a threshold value.

Embodiment 216. A non-volatile computer-readable medium for training a generative adversarial network (GAN), wherein the non-volatile computer-readable medium stores processor-executable instructions that, when executed by one or more processors, cause the device to: generate, in accordance with a GAN parameter set, a first simulated data set, containing simulated positive interactions of the polypeptide-MHC-I for the MHC allele; combining the first simulated dataset with positive real polypeptide-MHC-I interactions for the MHC allele and negative real polypeptide-MHC-I interactions for the MHC allele to create a GAN training dataset; obtaining information from the discriminator, the discriminator being configured to determine, according to a decision boundary, whether a positive interaction of the polypeptide-MHC-I for the MHC allele in the GAN training data set is positive or negative; an adjustment based on the accuracy of the information from the discriminator, one or more of the GAN parameter set, or the decision boundary; repeating steps a-d until the first stop criterion is met; using the GAN generator, according to the GAN parameter set, generating a second simulated dataset containing simulated positive polypeptide-MHC-I interactions for the MHC allele; combining the second simulated data set, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data for the MHC allele to create a CNN training data set; representing a CNN training dataset into a convolutional neural network (CNN); obtaining training information from the CNN, wherein the CNN is configured to determine the training information by classifying according to the polypeptide-MHC-I interaction parameter set of the CNN for the MHC allele in the training data set of the CNN as positive or negative; adjusting, based on the accuracy of the training information, one or more of the set of CNN parameters; repeating steps h-j until the second stop criterion is met; presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data to the CNN; obtaining information for training from the CNN, where the CNN is configured to determine the training information by classifying according to a set of CNN interaction parameters of the polypeptide-MHC-I for the MHC allele as positive or negative; and determining the accuracy of the training information, wherein when (if) the accuracy of the training information satisfies the third stopping criterion, deriving the GAN and CNN,

however, when (if) the accuracy of the training information does not satisfy the third stopping criterion, return to stage a.

Embodiment 217. The nonvolatile computer-readable medium of embodiment 216, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 218. The non-volatile computer-readable medium of embodiment 216, wherein the MHC allele is an HLA allele.

Embodiment 219. The nonvolatile computer-readable medium of embodiment 218, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 220. The nonvolatile computer-readable medium of Embodiment 218, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 221. The non-volatile computer-readable medium of embodiment 218, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 222. The nonvolatile computer-readable medium of embodiment 216, wherein the processor-executable instructions, when executed by one or more processors, further cause the one or more processors to: present a data set to a CNN, wherein the data set contains a plurality of candidate polypeptide-MHC-I interactions wherein the CNN is further configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified by the CNN as a positive polypeptide-MHC-I interaction.

Embodiment 223. The polypeptide obtained using the non-volatile computer-readable medium of Embodiment 222.

Embodiment 224. The nonvolatile computer-readable medium of Embodiment 222, wherein the polypeptide is a tumor-specific antigen.

Embodiment 225. The nonvolatile computer-readable medium of embodiment 222, wherein the polypeptide contains an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele.

Embodiment 226. The nonvolatile computer-readable medium of embodiment 216, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 227. The nonvolatile computer-readable medium of Embodiment 226, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 228. The nonvolatile computer-readable medium of embodiment 216, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to repeat steps a-d until the first stop criterion is met, further provide processor-executable instructions, which, when executed by one or more processors, causes one or more processors to evaluate the gradient descent expression for the GAN generator.

Embodiment 229. The nonvolatile computer-readable medium of embodiment 216, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to repeat steps a-d until the first stop criterion is met, further provide processor-executable instructions, which, when executed by one or more processors, causes one or more processors to perform the following: iteratively executes (e.g., optimizes) the GAN discriminator to increase the ability to assign high probability to positive real polypeptide-MHC-I interaction data, low probability to positive simulated polypeptide-MHC-I interaction data I and low probability of negative simulated polypeptide-MHC-I interaction data; and iteratively performing (eg, optimizing) the GAN generator to increase the possibility that positive simulated polypeptide-MHC-I interaction data will be highly valued.

Embodiment 230. The nonvolatile computer-readable medium of embodiment 216, wherein processor-executable instructions that, when executed by one or more processors, cause one or more processors to present data to a CNN for training the CNN, further provide processor-executable instructions that, when executed by one or more processors cause one or more processors to perform the following: perform a convolution procedure; performing a non-linearity procedure (eg, ReLu), performing a pooling or subsampling procedure; and performing a classification procedure (fully connected layer).

Embodiment 231. The non-volatile computer-readable medium of embodiment 216, wherein the GAN includes a deep convolutional GAN (DCGAN).

Embodiment 232. The nonvolatile computer-readable medium of embodiment 216, wherein the first stop criterion includes an estimate of a mean square error (MSE) function.

Embodiment 233. The nonvolatile computer-readable medium of embodiment 216, wherein the second stop criterion includes an estimate of a mean square error (MSE) function.

Embodiment 234. The nonvolatile computer-readable medium of embodiment 216, wherein the third stopping criterion includes an area under the curve (AUC) function estimate.

Embodiment 235. A non-volatile computer-readable medium for training a generative adversarial network (GAN), wherein the non-volatile computer-readable medium stores processor-executable instructions that, when executed by one or more processors, causes one or more processors to perform the following: learning a convolutional neural network (CNN) by the same means that the device according to option 83 implementation; presenting the dataset to a CNN, where the dataset comprises a plurality of candidate polypeptide-MHC-I interactions, wherein the CNN is configured to classify each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide associated with a candidate polypeptide-MHC-I interaction classified by the CNN as a positive polypeptide-MHC-I interaction.

Embodiment 236. The nonvolatile computer-readable medium of embodiment 235, wherein the CNN is trained based on one or more GAN parameters including one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size.

Embodiment 237. The non-volatile computer-readable medium of embodiment 236, wherein the HLA allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof.

Embodiment 238. The non-volatile computer-readable medium of Embodiment 236, wherein the HLA allele length is from about 8 to about 12 amino acids.

Embodiment 239. The non-volatile computer-readable medium of Embodiment 236, wherein the HLA allele length is from about 9 to about 11 amino acids.

Embodiment 240. The polypeptide obtained using the non-volatile computer-readable medium of Embodiment 235.

Embodiment 241. The nonvolatile computer-readable medium of Embodiment 235, wherein the polypeptide is a tumor-specific antigen.

Embodiment 242. The nonvolatile computer-readable medium of embodiment 235, wherein the polypeptide comprises an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected human leukocyte antigen (HLA) allele.

Embodiment 243. The nonvolatile computer-readable medium of embodiment 235, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with a selected allele.

Embodiment 244. The nonvolatile computer-readable medium of Embodiment 243, wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705, and combinations thereof.

Embodiment 245. The non-volatile computer-readable medium of embodiment 235, wherein the GAN includes a deep convolutional GAN (DCGAN).

Claims (51)

1. A method for training a generative adversarial network (GAN), which includes: a. obtaining positive and negative real data as a result of measuring the biological conditions of a person, animals, microorganisms, viruses or plants; b. generating, by the computing device with the GAN generator, more and more accurate positive simulated data until the GAN discriminator classifies the positive simulated data as positive; c. representing, by the computing device, positive simulated data, positive real data, and negative real data to a convolutional neural network (CNN) until the CNN classifies each type of data as positive or negative; d. representing, by means of the computing device, positive real data and negative real data CNN to generate prediction estimates, and e. determining by the computing device, based on the prediction estimates, whether the GAN is trained or not trained, and when the GAN is not trained, repeating steps b-d until a determination is made based on the prediction estimates that the GAN is trained. 2. The method of claim 1, wherein the positive simulated data, positive real data, and negative real data include biological data. 3. The method of claim 1, wherein positive simulated data includes positive simulated polypeptide-major histocompatibility complex class I (MHC-I) interaction data, positive real data includes positive real polypeptide-MHC-I interaction data, and negative real data includes negative real data of polypeptide-MHC-I interaction. 4. The method of claim 3, wherein generating increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies positive simulated polypeptide-MHC-I interaction data as real, comprises: e. generating, by means of the GAN generator, in accordance with the GAN parameter set, a first simulated dataset containing simulated positive polypeptide-MHC-I interactions for the MHC allele; f. combining the first simulated dataset with positive real polypeptide-MHC-I interactions for the MHC allele and negative real polypeptide-MHC-I interactions for the MHC allele to create a GAN training dataset; g. determining, with a discriminator according to a decision boundary, whether the corresponding polypeptide-MHC-I interaction for the MHC allele in the GAN training dataset is simulated positive, real positive, or real negative; h. an adjustment based on the accuracy of the determination by the discriminator, one or more of the GAN parameter set, or the decision boundary, and i. repeat steps e through h until the first stopping criterion is met. 5. The method of claim 4, wherein presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until until CNN classifies the relevant polypeptide-MHC-I interaction data as positive or negative, provides: j. generating by the GAN generator in accordance with the GAN parameter set a second simulated dataset containing simulated positive interactions of the polypeptide-MHC-I for the MHC allele; k. combining a second simulated dataset, positive real-life polypeptide-MHC-I interactions for the MHC allele, and negative real-life polypeptide-MHC-I interactions for the MHC allele to create a CNN training dataset; l. representing a CNN training dataset into a convolutional neural network (CNN); m. classifying by the CNN according to the CNN parameter set the corresponding polypeptide-MHC-I interaction for the MHC allele in the CNN training data set as positive or negative; n. an adjustment based on the classification accuracy by the CNN, one or more of the set of CNN parameters, and o. repeating steps l-n until the second stopping criterion is met. 6. The method of claim 5, wherein presenting positive real polypeptide-MHC-I interaction data and negative real polypeptide-MHC-I interaction data in a CNN to generate predictive scores includes: classification by CNN according to the CNN parameter set of the corresponding polypeptide-MHC-I interaction for the MHC allele as positive or negative. 7. The method of claim 6, wherein determining whether the GAN has been trained based on the prediction scores comprises determining the classification accuracy by the CNN, wherein when the classification accuracy satisfies the third stopping criterion, deriving the GAN and the CNN. 8. The method of claim 6, wherein determining, based on the prediction scores, whether the GAN has been trained comprises determining the classification accuracy by the CNN, wherein, when the classification accuracy does not satisfy the third stopping criterion, returning to step b. 9. The method of claim 4, wherein the GAN parameters include one or more of allele type, allele length, generation category, model complexity, learning rate, or batch size. 10. The method of claim 9, wherein the allele type includes one or more of HLA-A, HLA-B, HLA-C, or a subtype thereof. 11. The method of claim 9, wherein the allele length is from about 8 to about 12 amino acids. 12. The method of claim 11, wherein the allele length is from about 9 to about 11 amino acids. 13. The method according to p. 3, additionally providing: representing a dataset in a CNN, where the dataset contains a plurality of candidate polypeptide-MHC-I interactions; classifying by CNN each of the plurality of candidate polypeptide-MHC-I interactions as positive or negative polypeptide-MHC-I interactions; and synthesizing a polypeptide from a candidate polypeptide-MHC-I interaction classified as a positive polypeptide-MHC-I interaction. 14. The method of claim 13 wherein the polypeptide is a tumor-specific antigen. 15. The method of claim 13, wherein the polypeptide contains an amino acid sequence that specifically binds to the MHC-I protein encoded by the selected MHC allele. 16. The method of claim 3, wherein positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data are associated with the selected allele. 17. The method of claim 16 wherein the selected allele is selected from the group consisting of A0201, A0202, A0203, B2703, B2705 and combinations thereof. 18. The method of claim 3, wherein generating progressively more accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies the positive simulated polypeptide-MHC-I interaction data as positive, involves evaluating the gradient descent expression for the generator GAN. 19. The method of claim 3, wherein generating increasingly accurate positive simulated polypeptide-MHC-I interaction data until the GAN discriminator classifies the positive simulated polypeptide-MHC-I interaction data as positive, comprises: iteratively executing the GAN discriminator to increase the ability to assign high probability to positive real polypeptide-MHC-I interaction data, low probability to positive simulated polypeptide-MHC-I interaction data, and low probability to negative real polypeptide-MHC-I interaction data, and iterative execution of the GAN generator to increase the likelihood that positive simulated polypeptide-MHC-I interaction data will score high. 20. The method of claim 3, wherein presenting positive simulated polypeptide-MHC-I interaction data, positive real polypeptide-MHC-I interaction data, and negative real polypeptide-MHC-I interaction data to a convolutional neural network (CNN) until until CNN classifies the relevant polypeptide-MHC-I interaction data as positive or negative, provides: performing the convolution procedure; execution of the non-linearity procedure (ReLU); performing a merging or subsampling procedure, and execution of the classification procedure (fully connected layer). 21. The method of claim 1, wherein the GAN includes a deep convolutional GAN (DCGAN). 22. The method of claim 8, wherein the first stopping criterion includes a mean square error (MSE) function estimate, the second stopping criterion includes a mean square error (MSE) function estimate, and the third stopping criterion includes an area under the curve (AUC) function estimate. 23. The method of claim 3, wherein the predictive score is the probability that positive real polypeptide-MHC-I interaction data is classified as positive polypeptide-MHC-I interaction data. 24. The method of claim 1, wherein determining, based on the prediction scores, whether the GAN has been trained comprises comparing one or more prediction scores to a threshold value. 25. The method of claim 1, further comprising deriving GANs and CNNs.

Images