KR102547967B1 - Apparatus and method for generating tcr information corresponding to pmhc using artificial intelligence - Google Patents

Abstract

A method of training an artificial intelligence-based predictive model performed by a computing device is disclosed. The method comprises: acquiring a first training data set related to CDR3 of a T Cell Receptor (TCR) - the first training data set contains guide information to be used in learning a first artificial intelligence-based prediction model Including -, based on the guide information, pre-processing the first training data set for learning the first prediction model, and using the pre-processed first training data set, the first prediction model The method may include training the first prediction model to predict an amino acid sequence of CDR3 of a TCR corresponding to a Major Histocompatibility Complex (MHC) and a peptide.

Description

Method and apparatus for generating TCR information corresponding to pMHC using artificial intelligence technology

The present disclosure relates to artificial intelligence technology, and more specifically to analyzing the relationship between a peptide-major histocompatibility complex (pMHC) and T Cell Receptor (TCR) using artificial intelligence technology.

The major histocompatibility complex is a genetic locus that encodes 'MHC molecules' that function in the immune system. MHC molecules can exist in type 1 (class I) and type 2 (class II).

Immunopeptidome refers to a set of peptides expressed on the cell surface, and for example, the immunopeptidome may refer to a combination of MHC-related peptides.

Human Leukocyte Antigen (HLA) is a glycoprotein molecule produced by the human major histocompatibility complex (MHC) gene. HLA does not exist in mature erythrocytes, but is expressed in immature erythroblasts and is expressed on the surface of all tissue cells in the human body, including blood cells such as leukocytes and/or platelets. The MHC gene is present in all vertebrates, and the human MHC gene is referred to as the HLA gene, and the product expressed therefrom is referred to as HLA.

MHC genes are involved in self and non-self recognition, immune response to antigen stimulation, regulation of cellular and humoral immunity, and susceptibility to disease. HLA, a product of the MHC gene, is the second most important antigen next to the ABO blood type in survival of the transplanted organ in solid organ transplantation. HLA is known to play the most important role in transplantation success or failure in bone marrow transplantation. Therefore, recognizing the difference in HLA immunologically can be seen as the first step of the rejection action for the transplanted tissue. In addition, in transfusion therapy, HLA and antibodies play an important role in the occurrence of various side effects such as platelet transfusion refractory, febrile nonhemorrhagic transfusion side effects, acute lung injury, and graft-versus-host disease after transfusion.

HLA, like MHC, can be largely classified into Class I and Class II. Class I is classified into HLA-A, HLA-B, and HLA-C, and is expressed in most nucleated cells and platelets, and antigen recognition when cytotoxic T cells recognize and remove virus-infected cells or tumor cells. ) is essential for HLA Class II is classified into HLA-DR, HLA-DQ, and HLA-DP, and is expressed in B cells, monocytes, dendritic cells, and activated T cells. and induces a humoral immune response, and is known to be essential when recognizing antigens expressed in antigen-presenting cells. HLA is a gene that shows the largest polymorphism among genes possessed by humans, and there is a frequency difference between races and ethnic groups.

When peptides derived from proteins derived from infectious microorganisms or proteins specific to cancer cells bind to MHC and are presented on the cell surface, T cells recognize them and trigger an immune response to eliminate infected cells or cancer cells. As such, T cells are key regulators (players) that determine specific immune responses to foreign substances that do not exist in the normal human body. Therefore, the prediction of TCR (T Cell Receptor) that binds to pMHC can be used for the development of personalized vaccines for the prevention of infectious diseases or cancer.

Korean Registered Patent No. 10-2322832

The present disclosure has been made in response to the above background art, and is intended to more efficiently and/or more accurately predict or identify a TCR capable of binding to pMHC.

The technical problems of the present disclosure are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below.

According to one embodiment of the present disclosure, a method of training an artificial intelligence-based predictive model performed by a computing device is disclosed. The method includes: acquiring a first training data set related to CDR3 of a T Cell Receptor (TCR) - the first training data set includes guide information to be used in learning of a first prediction model based on artificial intelligence Including -, based on the guide information, pre-processing the first training data set for learning the first prediction model, and using the pre-processed first training data set, the first prediction model The method may include training the first artificial intelligence prediction model to predict an amino acid sequence of CDR3 of a TCR corresponding to a Major Histocompatibility Complex (MHC) and a peptide.

In one embodiment, the preprocessing may include applying a mask to a frame including a predetermined number of consecutive units included in the first training data set. . The length of the frame may correspond to the number of consecutive units included in the frame.

In one embodiment, the length of the frame may be determined based on a geometric distribution indicating a relationship between the length of the frame and the sampling probability.

In an embodiment, the length of the frame may be determined based on an inverse value of a geometric distribution indicating a relationship between the length of the frame and the sampling probability.

In one embodiment, the length of the frame may be determined based on a scheme in which the same sampling probability is applied to the lengths of a plurality of different frames.

In an embodiment, the preprocessing step includes: determining a position of a frame including a predetermined number of consecutive units in the first training data set based on the guide information; and determining a position of the determined frame. Based on the location, it may include applying a mask to the frame.

In one embodiment, the preprocessing includes a predetermined number of consecutive units in the first training data set based on the length of the V type, J type and amino acid sequence included in the guide information. The method may include determining a location of the frame, and applying a mask to the frame based on the determined location of the frame.

In one embodiment, the guide information is obtained from the output of a second prediction model based on artificial intelligence different from the first prediction model, and the second prediction model is prediction including the guide information based on peptides and MHC You can print the result.

In one embodiment, the step of training the first predictive model is substituting each of the units other than the masked single unit with an amino acid X within a frame composed of a predetermined number of consecutive units, and the substituted and training the first prediction model so that the first prediction model predicts the masked single unit by using amino acid X.

In one embodiment, position information including a distance from the masked single unit is assigned to each of the units other than the masked single unit in the frame, and the step of learning the first predictive model, The method may include training the first prediction model so that the first prediction model predicts the masked single unit based on the location information.

In one embodiment, a first loss function determined based on a difference between an output expression of a unit included in the preprocessed first training data set and a correct answer of the unit, a frame to which a mask is applied in the first training data set A second loss function determined based on the output representation of units before the start position of , the output representation of units after the end position of the frame, and the embedding position of each of the units in the frame, the preprocessed first learning The output representation of the masked single unit and the masked single unit according to a scheme in which each of the units other than the masked single unit is substituted with an amino acid X within a frame consisting of a predetermined number of consecutive units in the data set A third loss function determined based on the difference between the correct answers of , or a CDR3 sequence output through the first prediction model and a CDR3 sequence of the TCR corresponding to the actual peptide and MHC A fourth loss determined based on the difference between the correct answers and training the first predictive model based on at least two of the functions.

In one embodiment, the second loss function may be assigned a higher weight than the first loss function.

In one embodiment, the step of learning the first predictive model is: determining a CDR3 sequence that cannot be output from the peptide and MHC from the subject from the information of the peptide, MHC and corresponding TCR CDR3 from the subject and training the first prediction model by excluding the determined non-outputable CDR3 sequence from the first training data set.

In one embodiment, the CDR3 sequence of the TCR to be excluded from the first training data set may be determined based on data included in a biostatistical database.

In one embodiment, the CDR3 sequence of the TCR to be excluded from the first learning data set may be determined based on data obtained by sequencing the subject by molecular biological experiments.

In one embodiment, the step of applying a mask to the frame may include, when the first prediction model predicts a mask corresponding to each of the consecutive units in the frame, amino acids to the remaining mask other than the mask to be predicted It may include applying an amino acid X that reflects the factor of the species.

In one embodiment, the predictive model is learned by applying a mask to frames containing a predetermined number of contiguous units, and the predictive model is trained on the number of frames to which the mask is applied in the first training data set. Based on the start value corresponding to the unit before the start position and the end value corresponding to the unit after the end position of the frame, learning by matching the value corresponding to at least one unit included in the frame to which the mask is applied. It can be.

A computer program stored in a computer readable storage medium according to an embodiment is disclosed. The computer program, when executed by a computing device, causes the computing device to perform operations for learning an artificial intelligence-based predictive model, the operations comprising: obtaining a first training data set related to CDR3 of the TCR - the The first training data set includes guide information to be used in learning the first prediction model based on artificial intelligence - Based on the guide information, the first training data set is preprocessed for learning the first prediction model and, using the preprocessed first training data set, learning the first predictive model so that the first predictive model predicts the amino acid sequence of CDR3 of the TCR corresponding to the major histocompatibility complex and the peptide. can include

A computing device according to an embodiment is disclosed. The computing device may include at least one processor; and memory. The at least one processor: Obtaining a first training data set related to CDR3 of the TCR, wherein the first training data set includes guide information to be used in learning a first prediction model based on artificial intelligence. Based on the guide information, pre-processing the first training data set for learning the first predictive model, and using the pre-processed first training data set, the first predictive model is the main histocompatibility complex and An operation of training the first prediction model to predict the amino acid sequence of CDR3 of the TCR corresponding to the peptide may be performed.

The method and apparatus according to an embodiment of the present disclosure can more efficiently and/or more accurately predict or identify a TCR capable of binding to pMHC.

1 schematically illustrates a block configuration diagram of a computing device according to an embodiment of the present disclosure.
2 illustrates an exemplary structure of an artificial intelligence-based model according to an embodiment of the present disclosure.
3 exemplarily illustrates a method of learning a predictive model that generates a prediction result including a CDR3 sequence of a TCR according to an embodiment of the present disclosure.
4A exemplarily illustrates a method of training a predictive model that generates a predictive result including a CDR3 sequence of a TCR through a preprocessed training data set, according to an embodiment of the present disclosure.
4B exemplarily illustrates a method of training a predictive model that generates a predictive result including a CDR3 sequence of a TCR through a preprocessed training data set, according to an embodiment of the present disclosure.
5 exemplarily illustrates various masking schemes in a preprocessing process of a training data set according to an embodiment of the present disclosure.
6 exemplarily illustrates a method of generating a prediction result including the CDR3 sequence of the TCR through the first prediction model.
7 exemplarily illustrates a method of generating a prediction result comprising a CDR3 sequence of a TCR according to an embodiment of the present disclosure.
8 is a schematic diagram of a computing environment according to one embodiment of the present disclosure.

Various embodiments are described with reference to the drawings. In this specification, various descriptions are presented to provide an understanding of the present disclosure. Prior to describing specific details for the implementation of the present disclosure, it should be noted that configurations not directly related to the technical gist of the present disclosure have been omitted within the scope of not distracting from the technical gist of the present invention. In addition, the terms or words used in this specification and claims have meanings consistent with the technical idea of the present invention based on the principle that the inventor can define the concept of appropriate terms in order to best describe his/her invention. concept should be interpreted.

As used herein, the terms "component", "module", "system", "unit", and the like refer to a computer-related entity, hardware, firmware, software, a combination of software and hardware, or an execution of software, and are interchangeable. can possibly be used. For example, a component may be, but is not limited to, a procedure, processor, object, thread of execution, program, and/or computer running on a processor. For example, both an application running on a computing device and a computing device may be components. One or more components may reside within a processor and/or thread of execution. A component can be localized within a single computer. A component may be distributed between two or more computers. Also, these components can execute from various computer readable media having various data structures stored thereon. Components may be connected, for example, via signals with one or more packets of data (e.g., data and/or signals from one component interacting with another component in a local system, distributed system) to other systems and over a network such as the Internet. data being transmitted) may communicate via local and/or remote processes.

In addition, the term “or” is intended to mean an inclusive “or” rather than an exclusive “or”. That is, unless otherwise specified or clear from the context, “X employs A or B” is intended to mean one of the natural inclusive substitutions. That is, X uses A; X uses B; Or, if X uses both A and B, "X uses either A or B" may apply to either of these cases. Also, the term "and/or" as used herein should be understood to refer to and include all possible combinations of one or more of the listed related items.

Also, the terms "comprises" and/or "comprising" should be understood to mean that the features and/or components are present. However, it should be understood that the terms "comprises" and/or "comprising" do not exclude the presence or addition of one or more other features, elements, and/or groups thereof. Also, unless otherwise specified or where the context clearly indicates that a singular form is indicated, the singular in this specification and claims should generally be construed to mean "one or more".

In addition, the term "at least one of A or B" or "at least one of A and B" means "includes only A", "includes only B", "includes A and B in combination" should be interpreted as meaning

Those skilled in the art will further understand that the various illustrative logical components, blocks, modules, circuits, means, logics, and algorithms described in connection with the embodiments disclosed herein may be electronic hardware, computer software, or combinations of both. It should be recognized that it can be implemented as To clearly illustrate the interchangeability of hardware and software, various illustrative components, blocks, means, logics, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented in hardware or as software depends on the particular application and design constraints imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application. However, such implementation decisions should not be interpreted as causing a departure from the scope of this disclosure.

The description of the presented embodiments is provided to enable any person skilled in the art to use or practice the present invention. Various modifications to these embodiments will be apparent to those skilled in the art of this disclosure. The general principles defined herein may be applied to other embodiments without departing from the scope of this disclosure. Thus, the present invention is not limited to the embodiments presented herein. The present invention is to be accorded the widest scope consistent with the principles and novel features set forth herein.

In the present disclosure, terms expressed as Nth, such as first, second, or third, are used to distinguish at least one entity. For example, entities represented as first and second may be the same as or different from each other.

In the present disclosure, for convenience of description, human leukocyte antigen (HLA) is exemplarily used as an example for MHC. Therefore, the description of HLA or MHC used below is an example for expressing a description of MHC or HLA, and the scope of the present disclosure will be determined based on the content described in the claims, through examples of HLA. The scope of the right should not be construed as limited to HLA. As such, HLA and MHC in the present disclosure may be used interchangeably.

As used in the present disclosure, the term “human leukocyte antigen (HLA)” is a glycoprotein molecule produced by the human MHC gene, and is a gene that shows the largest polymorphism among genes possessed by humans. HLA typing, which determines the HLA type, can be used very actively in various fields such as organ transplantation, immunotherapy, disease-related research, paternity tests such as paternity, forensic use, and genetic research.

An HLA type in the present disclosure may include, for example, an HLA-A type, an HLA-B type, and/or an HLA-C type.

The conjugate of MHC and peptide in the present disclosure may mean a peptide antigen presenting complex through MHC class I molecules through processing through proteasome in antigen presenting cell (APC). The proteasome is composed of two units, LMP-2 and LMP-7 (low molecular weight polypeptide). These two proteasome units are located near the TAP-1 and TAP-2 genes in the MHC gene. The proteasome subunits are particularly important for the degradation of peptides bound to MHC I molecules. When cells are treated with the cytokine interferon gamma (IFN-γ), the expression of LMP-2 and LMP-7 can be induced. Expression of LMP-2 and LMP-7 results in a change in the substrate specificity of the proteasome, increasing its ability to degrade into peptides. IFN-γ increases the expression of proteins related to antigen presentation, such as MHC I and MECL-1, as well as LMP proteins, so that antigen presentation can be increased in antigen presenting cells.

MHC I molecules bind not only to peptide antigens degraded by the proteasome, but also to peptides produced by proteolytic enzymes present in the endoplasmic reticulum.

Tapasin serves as a bridge between TAP-1 (transporter associated with antigen processing) and MHC I, which has a stable tertiary structure in the endoplasmic reticulum (ER). It leaves the Shin and TAP proteins and becomes a complete peptide-MHC class I complex.

There are two types of T cell receptor (TCR), mainly composed of TCRα and TCRβ. The TCRα chain is located on chromosome 14, and the TCRβ chain is located at an independent locus on chromosome 7. Similar to the immunoglobulin (Ig) heavy chain, TCRβ is composed of a V gene segment, a D gene segment, a J gene segment, and a C gene segment, and the TCRα chain has a D gene segment similar to an Ig light chain. It is composed of V, J, and C gene segments.

The recombination process of the TCR is similar to that of the B cell receptor. The TCRα chain is encoded by the V, J and C gene segments, similar to the immunoglobulin light chain, and the TCRβ chain binds DJ to V-D-J, which is formed by V-D-J. formed through genetic recombination. The linkage of V-(D)-J is mediated by RAG-1 and RAG-2 (recombination-activating gene).

The hairpin structure is cleaved by an endonuclease, and a complementary P-nucleotide is added to the short strand of DNA formed at this time. In addition, 1 to 20 nucleotides may be randomly added to the cleavage ends, which are referred to as N-nucleotides, and this process is mediated by TdT (terminal deoxynucleotide transferase). P- and N-nucleotides increase the diversity of T-cell receptors.

T cells recognize only foreign antigens presented on the cell surface, and the activity of mature peripheral T cells is initiated by the interaction between antigens and peptides presented at the peptide binding cleft of TCR and MHC molecules.

In particular, most immune profiling focuses on the analysis of the CDR3 region within the TCR sequence. The CDR3 region is an important region involved in the interaction between antigen and receptor, and the most mutations are identified.

1 schematically illustrates a block configuration diagram of a computing device 100 according to one embodiment of the present disclosure.

Computing device 100 according to an embodiment of the present disclosure may include a processor 110 and a memory 130 .

The configuration of the computing device 100 shown in FIG. 1 is only a simplified example. In one embodiment of the present disclosure, the computing device 100 may include other components for performing a computing environment of the computing device 100, and only some of the disclosed components may constitute the computing device 100.

The computing device 100 in the present disclosure may mean any type of node constituting a system for implementing embodiments of the present disclosure. The computing device 100 may refer to any type of user terminal or any type of server. Components of the aforementioned computing device 100 are exemplary and some may be excluded or additional components may be included. For example, when the aforementioned computing device 100 includes a user terminal, an output unit (not shown) and an input unit (not shown) may be included within the scope of the computing device 100 .

The computing device 100 in the present disclosure may perform technical features according to embodiments of the present disclosure to be described later. For example, the computing device 100 may generate a prediction result including the CDR3 sequence of the TCR corresponding to the input data using an artificial intelligence-based prediction model using input data corresponding to MHC and peptide. For example, the computing device 100 may obtain information on MHC and peptides from a sample obtained from a subject, and generate a prediction result corresponding to CDR3 of the TCR based on the information on MHC and peptides.

In one embodiment of the present disclosure, the computing device 100 may obtain a result of base sequence analysis (eg, Next Generation Sequencing) from a server or an external entity. In another embodiment, the computing device 100 may perform nucleotide sequence analysis on genetic data (eg, DNA or RNA) obtained from a biological sample derived from a subject. The term used in the present disclosure, sequencing may be performed by any type of technique capable of analyzing the sequence of bases, for example, whole genome sequencing, whole exome It may include, but is not limited to, whole exome sequencing or whole transcriptome sequencing.

As used in the present disclosure, a subject may refer to a subject or individual for obtaining a biological sample containing a major histocompatibility complex (MHC), a peptide, and/or a complex thereof.

As used in the present disclosure, the term, sample, can be used without limitation as long as it is obtained from an individual or subject whose MHC type is to be determined, and for example, cells or tissues obtained by biopsy, blood, whole blood, serum, plasma, saliva, It may be cerebrospinal fluid, various secretions, urine and/or feces, and the like. Preferably, the sample may be selected from the group consisting of blood, plasma, serum, saliva, nasal fluid, sputum, ascites, vaginal secretion and/or urine, and more preferably blood, plasma or serum. The sample may be pre-treated prior to use in detection or diagnosis. For example, pretreatment methods may include homogenization, filtration, distillation, extraction, concentration, inactivation of interfering components, and/or addition of reagents, and the like. In the present disclosure, a biological sample may be tissue, cell, whole blood, and/or blood, but is not limited thereto.

In one embodiment, the processor 110 may include at least one core, and may include a central processing unit (CPU) or a general purpose graphics processing unit (GPGPU) of the computing device 100 . , a processor for data analysis and/or processing, such as a tensor processing unit (TPU).

The processor 110 may read the computer program stored in the memory 130 to generate a prediction result including the CDR3 sequence of the TCR according to an embodiment of the present disclosure.

According to an embodiment of the present disclosure, the processor 110 may perform an operation for learning a neural network. The processor 110 is used for neural network learning, such as processing input data for learning in deep learning (DL), extracting features from input data, calculating errors, and updating neural network weights using backpropagation. calculations can be performed. At least one of the CPU, GPGPU, and TPU of the processor 110 may process learning of the network function. For example, the CPU and GPGPU can process learning of network functions and data classification using network functions. In addition, in one embodiment of the present disclosure, learning of a network function and data classification using a network function may be processed by using processors of a plurality of computing devices together. In addition, a computer program executed in a computing device according to an embodiment of the present disclosure may be a CPU, GPGPU or TPU executable program.

Additionally, the processor 110 may typically handle overall operations of the computing device 100 . For example, the processor 110 processes data, information, signals, etc. input or output through components included in the computing device 100 or drives an application program stored in a storage unit, thereby providing appropriate information or information to the user. A function can be provided or processed.

According to one embodiment of the present disclosure, memory 130 may store any type of information generated or determined by processor 110 and any type of information received by computing device 100 . According to one embodiment of the present disclosure, the memory 130 may be a storage medium that stores computer software that causes the processor 110 to perform operations according to embodiments of the present disclosure. Accordingly, the memory 130 may refer to computer readable media for storing software codes necessary for performing embodiments of the present disclosure, data subject to execution of the codes, and results of execution of the codes.

According to an embodiment of the present disclosure, the memory 130 may refer to any type of storage medium. For example, the memory 130 may be a flash memory type, a hard disk type ), multimedia card micro type, card type memory (eg SD or XD memory, etc.), RAM (Random Access Memory, RAM), SRAM (Static Random Access Memory), ROM (Read-Only memory, ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, and optical disk. The computing device 100 may operate in relation to a web storage that performs a storage function of the memory 130 on the Internet. The description of the above memory is only an example, and the memory 130 used in the present disclosure is not limited to the above example.

The communication unit (not shown) in the present disclosure may be configured regardless of its communication mode, such as wired and wireless, and may be configured in various communication networks such as a personal area network (PAN) and a wide area network (WAN). can be configured. In addition, the network unit 150 may operate based on the known World Wide Web (WWW), and a wireless transmission technology used for short-range communication such as Infrared Data Association (IrDA) or Bluetooth. can also be used.

The computing device 100 in the present disclosure may include any type of user terminal and/or any type of server. Accordingly, embodiments of the present disclosure may be performed by a server and/or a user terminal.

A user terminal may include any type of terminal capable of interacting with a server or other computing device. User terminals include, for example, mobile phones, smart phones, laptop computers, personal digital assistants (PDAs), slate PCs, tablet PCs, and ultrabooks. can include

A server may include any type of computing system or computing device, such as, for example, microprocessors, mainframe computers, digital processors, portable devices and device controllers, and the like.

In a further embodiment, the aforementioned server may refer to an entity that stores and manages TCR information, immunopeptidome information, peptide sequence information, base sequence information, or gene information. The server is used to store immunopeptidome information, peptide sequence information, positional amino acid identifier information, nucleotide sequence information, gene information or database (eg, McPAS, TCR3F, huARdb, VDJdb, IMGT) reliability information, etc. It may include a storage unit (not shown), and the storage unit may be included in a server or may exist under the management of a server. As another example, the storage unit may exist outside the server and may be implemented in a form capable of communicating with the server. In this case, the storage unit may be managed and controlled by another external server different from the server.

2 illustrates an exemplary structure of an artificial intelligence-based model according to an embodiment of the present disclosure.

Throughout this specification, the terms predictive model, artificial intelligence-based predictive model, artificial intelligence model, artificial intelligence-based model, computational model, neural network, network function, and neural network may be used interchangeably.

A neural network may consist of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network includes one or more nodes. Nodes (or neurons) constituting neural networks may be interconnected by one or more links.

In a neural network, one or more nodes connected through a link may form a relative relationship of an input node and an output node. The concept of an input node and an output node is relative, and any node in an output node relationship with one node may have an input node relationship with another node, and vice versa. As described above, an input node to output node relationship may be created around a link. More than one output node can be connected to one input node through a link, and vice versa.

In a relationship between an input node and an output node connected through one link, the value of data of the output node may be determined based on data input to the input node. Here, a link interconnecting an input node and an output node may have a weight. The weight may be variable, and may be changed by a user or an algorithm in order to perform a function desired by the neural network. For example, when one or more input nodes are interconnected by respective links to one output node, the output node is set to a link corresponding to values input to input nodes connected to the output node and respective input nodes. An output node value may be determined based on the weight.

As described above, in the neural network, one or more nodes are interconnected through one or more links to form an input node and output node relationship in the neural network. Characteristics of the neural network may be determined according to the number of nodes and links in the neural network, an association between the nodes and links, and a weight value assigned to each link. For example, when there are two neural networks having the same number of nodes and links and different weight values of the links, the two neural networks may be recognized as different from each other.

A neural network may be composed of a set of one or more nodes. A subset of nodes constituting a neural network may constitute a layer. Some of the nodes constituting the neural network may form one layer based on distances from the first input node. For example, a set of nodes having a distance of n from the first input node may constitute n layers. The distance from the first input node may be defined by the minimum number of links that must be passed through to reach the corresponding node from the first input node. However, the definition of such a layer is arbitrary for explanation, and the order of a layer in a neural network may be defined in a method different from the above. For example, a layer of nodes may be defined by a distance from a final output node.

In one embodiment of the present disclosure, a set of neurons or nodes may be defined as a layer.

An initial input node may refer to one or more nodes to which data is directly input without going through a link in relation to other nodes among nodes in the neural network. Alternatively, in a relationship between nodes based on a link in a neural network, it may mean nodes that do not have other input nodes connected by a link. Similarly, the final output node may refer to one or more nodes that do not have an output node in relation to other nodes among nodes in the neural network. Also, the hidden node may refer to nodes constituting the neural network other than the first input node and the last output node.

In the neural network according to an embodiment of the present disclosure, the number of nodes in the input layer may be the same as the number of nodes in the output layer, and the number of nodes decreases and then increases again as the number of nodes progresses from the input layer to the hidden layer. can In addition, the neural network according to another embodiment of the present disclosure may be a neural network in which the number of nodes of the input layer may be less than the number of nodes of the output layer and the number of nodes decreases as the number of nodes increases from the input layer to the hidden layer. there is. In addition, the neural network according to another embodiment of the present disclosure is a neural network in which the number of nodes in the input layer may be greater than the number of nodes in the output layer, and the number of nodes increases as the number of nodes increases from the input layer to the hidden layer. can A neural network according to another embodiment of the present disclosure may be a neural network in the form of a combination of the aforementioned neural networks.

A deep neural network (DNN) may refer to a neural network including a plurality of hidden layers in addition to an input layer and an output layer. Deep neural networks can reveal latent structures in data. That is, the structure of a photograph, text, video, audio, protein sequence structure, gene sequence structure, peptide sequence structure, amino acid sequence structure, music's latent structure (e.g. what objects are in a photograph, text content and emotion). what it is, what the content and emotion of the voice are, etc.), the binding affinity between TCR and pMHC, and/or the binding affinity between peptide and MHC. Deep neural networks include convolutional neural networks (CNNs), recurrent neural networks (RNNs), auto encoders, generative adversarial networks (GANs), restricted Boltzmann machines ( It may include a restricted boltzmann machine (RBM), a deep belief network (DBN), a Q network, a U network, a Siamese network, and the like. The description of the deep neural network described above is only an example, and the present disclosure is not limited thereto.

As an example, the artificial intelligence-based predictive model of the present disclosure, the artificial intelligence-based predictive model, RNN (Recurrent Neural Network), LSTM (Long Short Term Memory) network, BERT (Bidirectional Encoder Representations from Transformers), or SpanBERT (Span Bidirectional Encoder Representations from Transformers).

The artificial intelligence-based predictive model of the present disclosure may be represented by a network structure of any of the foregoing structures including an input layer, a hidden layer, and an output layer.

A neural network that can be used in the artificial intelligence-based model of the present disclosure is at least one of supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning. can be learned in this way. Learning of the neural network may be a process of applying knowledge for the neural network to perform a specific operation to the neural network. For example, the predictive model according to an embodiment of the present disclosure applies a mask to at least some of the frames or units constituting amino acid sequences, and then semi-supervised learning to fit the masked frames or units. (semi-supervised learning) method.

A neural network can be trained in a way that minimizes output errors. In the learning of the neural network, the learning data is repeatedly input into the neural network, the output of the neural network for the training data and the error of the target are calculated, and the error of the neural network is transferred from the output layer of the neural network to the input layer in the direction of reducing the error. It is a process of updating the weight of each node of the neural network by backpropagating in the same direction. In the case of supervised learning, each learning data is labeled with the correct answer (ie, labeled learning data), and in the case of unsupervised learning, the correct answer may not be labeled in each learning data. That is, for example, learning data in the case of supervised learning related to data classification may be data in which each learning data is labeled with a category. Labeled training data is input to a neural network, and an error may be calculated by comparing an output (category) of the neural network and a label of the training data. As another example, in the case of unsupervised learning for data classification, an error may be calculated by comparing input learning data with a neural network output. The calculated error is back-propagated in a reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back-propagation. The amount of change in the connection weight of each updated node may be determined according to a learning rate. The neural network's computation of input data and backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently according to the number of iterations of the learning cycle of the neural network. For example, a high learning rate may be used in the early stage of neural network training to increase efficiency by allowing the neural network to quickly obtain a certain level of performance, and a low learning rate may be used in the late stage to increase accuracy.

In neural network learning, generally, training data can be a subset of real data (ie, data to be processed using the trained neural network), and therefore, errors for training data are reduced, but errors for real data are reduced. There may be incremental learning cycles. Overfitting is a phenomenon in which errors for actual data increase due to excessive learning on training data. For example, a phenomenon in which a neural network that has learned a cat by showing a yellow cat does not recognize that it is a cat when it sees a cat other than yellow may be a type of overfitting. Overfitting can act as a cause of increasing the error of machine learning algorithms. Various optimization methods can be used to prevent such overfitting. In order to prevent overfitting, methods such as increasing the training data, regularization, inactivating some nodes in the network during learning, and using a batch normalization layer are methods. can be applied

A computer readable medium storing a data structure according to an embodiment of the present disclosure is disclosed. The above-described data structure may be stored in a storage unit in the present disclosure, executed by a processor, and transmitted and received by a communication unit.

Data structure can refer to the organization, management, and storage of data that enables efficient access and modification of data. Data structure may refer to the organization of data to solve a specific problem (eg, data retrieval, data storage, data modification in the shortest time). A data structure may be defined as a physical or logical relationship between data elements designed to support a specific data processing function. A logical relationship between data elements may include a connection relationship between user-defined data elements. A physical relationship between data elements may include an actual relationship between data elements physically stored in a computer-readable storage medium (eg, a persistent storage device). The data structure may specifically include a set of data, a relationship between data, and a function or command applicable to the data. Through an effectively designed data structure, a computing device can perform calculations while using minimal resources of the computing device. Specifically, the computing device can increase the efficiency of operation, reading, insertion, deletion, comparison, exchange, and search through an effectively designed data structure.

The data structure can be divided into a linear data structure and a non-linear data structure according to the shape of the data structure. A linear data structure may be a structure in which only one data is connected after one data. Linear data structures may include lists, stacks, queues, and decks. A list may refer to a series of data sets in which order exists internally. The list may include a linked list. A linked list may be a data structure in which data are connected in such a way that each data is connected in a single line with a pointer. In a linked list, a pointer can contain information about connection to the next or previous data. A linked list can be expressed as a singly linked list, a doubly linked list, or a circular linked list depending on the form. A stack can be a data enumeration structure that allows limited access to data. A stack can be a linear data structure in which data can be processed (eg, inserted or deleted) at only one end of the data structure. The data stored in the stack may be a LIFO-Last in First Out (Last in First Out) data structure. A queue is a data listing structure that allows limited access to data, and unlike a stack, it can be a data structure (FIFO-First in First Out) in which data stored later comes out later. A deck can be a data structure that can handle data from either end of the data structure.

The nonlinear data structure may be a structure in which a plurality of data are connected after one data. The non-linear data structure may include a graph data structure. A graph data structure can be defined as a vertex and an edge, and an edge can include a line connecting two different vertices. A graph data structure may include a tree data structure. The tree data structure may be a data structure in which one path connects two different vertices among a plurality of vertices included in the tree. That is, it may be a data structure that does not form a loop in a graph data structure.

Throughout this specification, artificial intelligence-based models, computational models, neural networks, network functions, and neural networks may be used interchangeably. Hereinafter, a neural network is unified and described. The data structure may include a neural network. And the data structure including the neural network may be stored in a computer readable medium. The data structure including the neural network may also include preprocessed data for processing by the neural network, data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation function associated with each node or layer of the neural network, and neural network It may include a loss function for learning of . A data structure including a neural network may include any of the components described above. That is, the data structure including the neural network includes preprocessed data for processing by the neural network, data input to the neural network, weights of the neural network, hyperparameters of the neural network, data obtained from the neural network, activation function associated with each node or layer of the neural network, and neural network. It may be configured to include all or any combination thereof, such as a loss function for learning of . In addition to the foregoing configurations, the data structure comprising the neural network may include any other information that determines the characteristics of the neural network. In addition, the data structure may include all types of data used or generated in the computational process of the neural network, but is not limited to the above. A computer readable medium may include a computer readable recording medium and/or a computer readable transmission medium. A neural network may consist of a set of interconnected computational units, which may generally be referred to as nodes. These nodes may also be referred to as neurons. A neural network includes one or more nodes.

The data structure may include data input to the neural network. A data structure including data input to the neural network may be stored in a computer readable medium. Data input to the neural network may include training data input during a neural network learning process and/or input data input to a neural network that has been trained. Data input to the neural network may include pre-processed data and/or data subject to pre-processing. Pre-processing may include a data processing process for inputting data to a neural network. Accordingly, the data structure may include data subject to pre-processing and data generated by pre-processing. The foregoing data structure is only an example, and the present disclosure is not limited thereto.

The data structure may include the weights of the neural network. (In this specification, weights and parameters may be used in the same meaning.) Also, a data structure including weights of a neural network may be stored in a computer readable medium. A neural network may include a plurality of weights. The weight may be variable, and may be changed by a user or an algorithm in order to perform a function desired by the neural network. For example, when one or more input nodes are interconnected by respective links to one output node, the output node is set to a link corresponding to values input to input nodes connected to the output node and respective input nodes. A data value output from an output node may be determined based on the weight. The foregoing data structure is only an example, and the present disclosure is not limited thereto.

As a non-limiting example, the weights may include weights that are varied during neural network training and/or weights for which neural network training has been completed. The variable weight in the neural network learning process may include a weight at the time the learning cycle starts and/or a variable weight during the learning cycle. The weights for which neural network learning has been completed may include weights for which learning cycles have been completed. Accordingly, the data structure including the weights of the neural network may include a data structure including weights that are variable during the neural network learning process and/or weights for which neural network learning is completed. Therefore, it is assumed that the above-described weights and/or combinations of weights are included in the data structure including the weights of the neural network. The foregoing data structure is only an example, and the present disclosure is not limited thereto.

The data structure including the weights of the neural network may be stored in a computer readable storage medium (eg, a memory or a hard disk) after going through a serialization process. Serialization can be the process of converting a data structure into a form that can be stored on the same or another computing device and later reconstructed and used. A computing device may serialize data structures to transmit and receive data over a network. The data structure including the weights of the serialized neural network may be reconstructed on the same computing device or another computing device through deserialization. The data structure including the weights of the neural network is not limited to serialization. Furthermore, the data structure including the weights of the neural network is a data structure for increasing the efficiency of operation while minimizing the resource of the computing device (for example, B-Tree, R-Tree, Trie, m-way search tree in a nonlinear data structure) , AVL tree, Red-Black Tree). The foregoing is only an example, and the present disclosure is not limited thereto.

The data structure may include hyper-parameters of the neural network. Also, the data structure including the hyperparameters of the neural network may be stored in a computer readable medium. A hyperparameter may be a variable variable by a user. Hyperparameters include, for example, learning rate, cost function, number of learning cycle iterations, weight initialization (eg, setting the range of weight values to be targeted for weight initialization), hidden unit number (eg, the number of hidden layers and the number of nodes in the hidden layer). The foregoing data structure is only an example, and the present disclosure is not limited thereto.

A transformer may be considered as a network function for a predictive model according to an embodiment of the present disclosure. As an example, the predictive model may be operated based on a transformer. Such a predictive model may be operated using, for example, a recurrent neural network to which an attention algorithm is applied or a transformer to which an attention algorithm is applied.

In one embodiment, a transformer may consist of an encoder that encodes the embedded data and a decoder that decodes the encoded data. The transformer may have a structure that receives a series of data and outputs a series of data of different types through encoding and decoding steps. In one embodiment, the series of data can be processed into a form operable by a transformer. A process of processing a series of data into a form in which a transformer can operate may include an embedding process. Expressions such as data token, embedding vector, and embedding token may refer to embedded data in a form that can be processed by a transformer.

In order for the transformer to encode and decode a series of data, encoders and decoders within the transformer may be processed using an attention algorithm. The Attention Algorithm is to obtain the similarity of one or more keys for a given query, reflect the given similarity to each key and corresponding value, and then reflect the similarity to the value ( Values) may mean an algorithm that calculates an attention value by weighting.

Depending on how to set the query, key, and value, various types of attention algorithms can be classified. For example, when attention is obtained by setting the same query, key, and value, this may mean a self-attention algorithm. In order to process a series of input data in parallel, when the dimension of the embedding vector is reduced and individual attention heads are obtained for each divided embedding vector to obtain attention, this means a multi-head attention algorithm. can do.

In one embodiment, a transformer may consist of modules that perform a plurality of multi-head self-attention algorithms or multi-head encoder-decoder algorithms. In one embodiment, the transformer may also include additional elements other than the attention algorithm, such as an embedding layer, a normalization layer, and a softmax layer. Methods for constructing transformers using the attention algorithm may include methods disclosed in Vaswani et al., Attention Is All You Need, 2017 NIPS, which is incorporated herein by reference.

Transformers can be applied to various data domains such as embedded natural language, embedded sequence information, segmented image data, and audio waveforms to convert a series of input data into a series of output data. In order to convert data having various data domains into a series of data that can be input to the transformer, the transformer can embed the data. Transformers can process additional data representing relative positional or phase relationships between a set of input data. Alternatively, a series of input data may be embedded by additionally reflecting vectors representing a relative positional relationship or phase relationship between input data to the series of input data. In one example, the relative positional relationship between a series of input data may include, but is not limited to, word order in a natural language sentence, relative positional relationship of each segmented image, temporal sequence of segmented audio waveforms, and the like. . A process of adding information representing a relative positional relationship or phase relationship between a series of input data may be referred to as positional encoding.

One example of how to embed data and transform it into a transformer is disclosed in Dosovitskiy, et al., AN IMAGE IS WORTH 16X16 WORDS: TRANSFORMERS FOR IMAGE RECOGNITION AT SCALE, which document is incorporated herein by reference.

3 exemplarily illustrates a method of learning a predictive model that generates a prediction result including a CDR3 sequence of a TCR according to an embodiment of the present disclosure.

In one embodiment, the steps shown in FIG. 3 may be performed by computing device 100 . In a further embodiment, the steps in FIG. 3 may be implemented by a plurality of entities, such that some of the steps shown in FIG. 3 are performed in a user terminal and others are performed in a server.

In one embodiment, the computing device 100 may obtain a first training data set corresponding to CDR3 of the TCR (S310).

As an example, the first training data set may include guide information to be used in learning the artificial intelligence-based first prediction model.

In an embodiment, the guide information may refer to data used to enable the first prediction model to output highly accurate data in a learning process and/or an inference process of the first prediction model. For example, guide information may include V type, J type, or the length of an amino acid sequence.

For example, the guide information may include an amino acid sequence representing an amino acid sequence of MHC that interacts with CDR3 of the TCR, an amino acid sequence of a peptide, or an amino acid sequence representing an MHC-peptide complex.

As another example, the guide information may include a form of V-type and J-type genetic nucleic sequences.

As another example, the guide information may include multiple sequence alignment (MSA) based on evolutionary genome data.

As another example, guide information may correspond to output data of a second prediction model different from the first prediction model. In this example, guide information may be generated by a second predictive model that takes peptides and/or MHCs as inputs.

This guide information can be used in the process of pre-processing the training data set input to the first prediction model.

In one example, an amino acid sequence may include a group of identifiers representing, for example, amino acids.

In one embodiment, the first training data set may be obtained from public databases and/or experimental result data.

In an additional embodiment, the first training data set or input data may include input data to which Blosum encoding or one-hot encoding is applied to an amino acid sequence.

In a further embodiment, the first training data set or input data includes a first feature indicating polarity between amino acids, a second feature indicating the size of amino acids, a third feature indicating whether the amino acids are hydrophobic or hydrophilic, and the presence of charge on amino acids. It may further include at least one of the fourth characteristic indicating whether the amino acid is aromatic or the fifth characteristic indicating whether the amino acid is aromatic or aliphatic.

In an embodiment, the computing device 100 may pre-process the first training data set for learning the first prediction model based on the guide information (S320).

As an example, preprocessing for learning of a predictive model may be performed through guide information generated through a second predictive model operating with HLA and peptides as input data.

For example, the computing device 100 may determine a location of a region to be masked in the first training data set based on the guide information. For example, a region to be masked may include a plurality of contiguous amino acids or a plurality of contiguous sets of amino acids. The computing device 100 may enable the first prediction model to perform masking-based learning based on the determined location.

The process of pre-processing the training data set will be described later with reference to FIGS. 4 and 6 .

The computing device 100 may train the first prediction model to predict the CDR3 amino acid sequence of the TCR corresponding to the MHC and the peptide using the preprocessed first training data set (S330).

Figure 4a is a unit constituting the CDR3 of the TCR included in the training data set according to an embodiment of the present disclosure, training a prediction model that generates a prediction result including the CDR3 sequence of the TCR through a preprocessed training data set The method is exemplarily shown.

In one embodiment, FIG. 4A illustratively illustrates a method for training a first predictive model including boundary task processing of a frame.

In one embodiment, the training method includes the steps of embedding units 410, 420, 430 constituting part of the CDR3 sequence of the preprocessed TCR, and the units constituting part of the CDR3 sequence of the TCR (410, 420, 430) may include embedding location data 440 reflecting location information in the CDR3 sequence.

In one embodiment, unit embedding may be included in the process of training the first prediction model. Through unit embedding, a first training data set may be formed by receiving m vectors of n dimensions for amino acids or units constituting the CDR3 unit of the TCR.

In one embodiment, position embedding may be included in the process of training the first prediction model. Location embedding may refer to an operation for reflecting location information about units constituting the CDR3 unit of the TCR. For example, location embedding may be performed in a separate embedding layer for reflecting location information about a unit.

For example, if the number of CDR3 units of the TCR has a length of 6, positional embedding may train the first prediction model with 6 positional embedding vectors.

In one embodiment, positional embedding may be performed based on the maximum length of the CDR3 amino acid sequence of the TCR. For example, the number of positional embedding vectors may be determined based on the maximum length of the amino acid sequence of CDR3. The length of the amino acid sequence of CDR3 usually consists of 11 to 18 amino acids, but is not limited thereto.

For example, the computing device 100 includes the embedding of the units 410, 420, and 430 constituting the CDR3 sequence of the preprocessed TCR and the location data corresponding to each of the units 410 of a portion of the CDR3 sequence N-terminus. 440 and the location data 440 corresponding to each of the units 420 of a portion of the CDR3 sequence C-terminus may be linked. Prediction data 460 may be generated in the first prediction model based on interworking between the unit embedding and the location data 440 . For example, location data 440 may be embedded to identify a location for each of the units 410, 420, and 430 of CDR3 in which the unit is embedded. As a result of location embedding, location data may be assigned to each of the unit-embedded units. As such, unit embedding and positional embedding may be performed for units corresponding to each other, and input data according to a result of applying both the positional data 440 embedding and the unit embedding may be input to the encoder. The encoder may generate output data corresponding to each of the units based on the input data.

In one embodiment, a unit constituting a part of the CDR3 sequence of the TCR may mean one or more consecutive amino acids in the first training data. In one example, a unit may represent one amino acid. As an example, a unit may represent a token comprising a plurality of amino acids.

For example, a unit constituting a part of the CDR3 sequence of the TCR may be an amino acid sequence or a single amino acid such as CASS, CAS, ASS, CA, AS, SS, C, A, and S.

For example, a minimum unit embedded in FIG. 4A or a minimum unit output by the encoder 450 may correspond to a unit.

In one embodiment, the frame 430 may refer to a plurality of consecutive units masked in the training data set.

As an example, the frame 430 may refer to continuous units masked in the middle of a CDR3 sequence consisting of 10 to 20 amino acids in the CDR3 sequence of the TCR of the first training data set.

As an example, the frame 430 may mean a portion in which consecutive units [CA] [SSR] [T] [RK] [Q] [AV] masked in the CDR3 sequence of the TCR of the first training data set are masked. can

In one embodiment, pre-processing the CDR3 sequence of the TCR may include a portion 410 of the N-terminus of the CDR3 sequence, a portion 420 of the C-terminus of the CDR3 sequence, and a portion in the middle of the CDR3 sequence to be masked. . In the drawings of the present disclosure, a portion in the middle of the CDR3 sequence is exemplified as a frame 430 to be masked, but according to an embodiment, consecutive units at any position constituting the CDR3 sequence are framed to be masked ( 430) can be formed.

In one embodiment, the first predictive model is the units constituting the CDR3 of the TCR processed through the encoder 450, and the location embedding data of the units are prediction data corresponding to a portion 430 of the CDR3 sequence to be masked (460).

In one embodiment, the first predictive model comprises units 412 of some of the units 410 of the CDR3 sequence N-terminal portion 410 and units 420 of the CDR3 sequence C-terminal portion 420 constituting the masking region boundary. Based on some of the units 422, prediction data 460 may be generated.

In one embodiment, the first prediction model includes positional data 440 corresponding to each of the units 410 of a portion of the CDR3 sequence N-terminus constituting the boundary of the masking region and units of a portion of the CDR3 sequence C-terminus ( 420) Prediction data 460 may be generated based on the location data 440 corresponding to each.

For example, amino acid R and amino acid S among the CDR3 sequence N-terminal units 410 have location information P ₃ and P ₄ , respectively, and amino acid E and amino acid V among the CDR3 sequence C-terminal units 420 are Each may have location information P ₉ and P ₁₀ .

In a further example, amino acid R and its corresponding location information P ₃ are linked to each other, amino acid S and its corresponding location information P ₄ are linked to each other, amino acid E and its corresponding location information P ₉ are linked to each other, and The amino acid V and the positional information P ₁₀ corresponding thereto may be interlocked and used to predict amino acid sequences of units in the masked frame 430 of the first training data set.

Figure 4b is a unit constituting the CDR3 of the TCR included in the training data set according to an embodiment of the present disclosure, training a prediction model that generates a prediction result including the CDR3 sequence of the TCR through a preprocessed training data set The method is exemplarily shown.

In one embodiment, FIG. 4B illustratively illustrates a method for training the first predictive model including frame boundary task processing.

In one embodiment, in the training process, the computing device 100 embeds the units 410, 420, 430 constituting the CDR3 sequence of the preprocessed TCR, and the units 410 constituting the CDR3 sequence of the TCR. , 420, 430) may embed positional data 440 reflecting positional information within the CDR3 sequence.

In one embodiment, pretreatment of the CDR3 sequence of the TCR comprises units 410 of a portion of the N-terminus of the CDR3 sequence, units 420 of a portion of the C-terminus of the CDR3 sequence, and a portion in the middle of the CDR3 sequence being masked. It may be based on guide information output from the second prediction model for the units 430 .

In one embodiment, based on the units constituting the CDR3 of the TCR processed through the encoder 450 and the position embedding data of the units, the first prediction model is for a portion 430 in the middle of the CDR3 sequence that is masked. One or more predictive data 460 may be generated.

In one embodiment, the first prediction model is a part of the unit 412 of the output representation corresponding to the units 410 of the part of the CDR3 sequence N-terminus constituting the boundary of the masking region and the part of the C-terminus of the CDR3 sequence Prediction data 460 may be generated based on units 422 of some of the output representations corresponding to units 420 .

In one embodiment, the first predictive model is the position data 440 corresponding to each of the units of the CDR3 sequence N-terminal portion 410 constituting the masking region boundary and the CDR3 sequence C-terminal portion 420 of Based on the location data 440 corresponding to each unit, prediction data 460 may be generated.

For example, the processing in which the prediction data 460 is generated in the first prediction model includes the embedding of units 410, 420, and 430 constituting the CDR3 sequence of the preprocessed TCR and a portion 410 of the N-terminus of the CDR3 sequence. The location data 440 corresponding to each of the units and the location data 440 corresponding to each of the units of the CDR3 sequence C-terminal portion 420 may be interlocked.

In one embodiment, the first predictive model replaces each of the units other than a single masked unit with an amino acid X within a frame consisting of a predetermined number of consecutive units, and uses the substituted amino acid X to The first prediction model may be trained to predict a single masked unit.

For example, when a plurality of masked units exist for one training data, when an artificial intelligence-based prediction model predicts one mask among the plurality of masked units, the average value of amino acids in the mask of another unit or all Amino acid X representing amino acids can be applied.

In one embodiment, the first predictive model may be operated using one loss function or a plurality of loss functions. Four loss functions are presented as examples of a plurality of loss functions according to an embodiment of the present disclosure.

In one embodiment, a first loss function determined based on a difference between an output representation of a unit included in the preprocessed first training data set and a correct answer of the unit may be considered. In one embodiment, an output representation of a unit before a start position of a frame to which a mask is applied in a first training data set, an output representation of a unit after an end position of the frame, and an embedding position of each of the units in the frame A second loss function determined based on may be considered. In one embodiment, the masked single unit according to a method of substituting each of the units other than the masked single unit with amino acid X in a frame composed of a predetermined number of consecutive units in the preprocessed first training data set A third loss function may be considered that is determined based on the difference between the outputted expression of <RTI ID=0.0>of</RTI> In one embodiment, a fourth loss function determined based on the difference between the CDR3 sequence output through the first prediction model and the correct answer of the CDR3 sequence of the TCR corresponding to the actual peptide and MHC may be considered.

In one embodiment, the first predictive model of the present disclosure may be trained using different loss functions. As an example, the first prediction model may be learned based on at least two of the aforementioned loss functions.

For example, a first loss function determined based on a difference between an output expression of a unit included in the training data set and a correct answer of the unit, and an output of a unit before the start position of a frame to which a mask is applied in the training data set The combined loss function based on the representation, the output representation of units after the end position of the frame, and the second loss function determined based on the embedding position of each of the units in the frame is, for example, the first loss function and It can be defined by the summation of the second loss function.

As a further example, a first loss function determined based on a difference between an output representation of a unit included in the training data set and a correct answer of the unit, and within a frame consisting of a predetermined number of consecutive units in the training data set Combined based on the third loss function determined based on the difference between the output expression of the masked single unit and the correct answer of the masked single unit according to the method of substituting each of the units other than the single masked unit with amino acid X The loss function may be defined by, for example, the sum of the first loss function and the third loss function.

As a further example, based on the output representation of the unit before the start position of the frame to which the mask is applied in the training data set, the output representation of the unit after the end position of the frame, and the embedding position of each of the units in the frame Output of a single unit masked according to the determined second loss function and a method of substituting each of the units other than the single unit masked with amino acid X in a frame composed of a predetermined number of consecutive units in the training data set A combined loss function based on a third loss function determined based on the difference between the correct answer of the masked expression and the masked single unit may be defined, for example, by the sum of the second loss function and the third loss function.

As a further example, the output expression of a single unit masked according to a method of substituting each of the units other than the single unit masked with amino acid X in a frame consisting of a predetermined number of consecutive units in the training data set and the above A third loss function determined based on the difference between the correct answers of the masked single unit, and a fourth determined based on the difference between the correct answer between the CDR3 sequence output through the prediction model and the correct answer of the CDR3 sequence of the TCR corresponding to the actual peptide and MHC A combined loss function based on the loss function may be defined, for example, by the summation of the third loss function and the fourth loss function.

In one embodiment, based on the output representation of units before the start position of the frame to which the mask is applied, the output representation of units after the end position of the frame, and the embedding position of each of the units in the frame in the training data set. A higher weight than other loss functions other than the second loss function may be assigned to the second loss function determined by the above process.

5 is one of preprocessing of a learning data set according to an embodiment of the present disclosure, and shows masking schemes by way of example.

5 exemplarily suggests various methods for determining the length of a frame to be masked in a learning process.

In one embodiment, the computing device 100 may apply a mask to a frame including a predetermined number of contiguous units included in the first training data set. The length of a frame may correspond to the number of consecutive units included in the frame.

(a) in FIG. 5 exemplarily shows that the length of a frame to be masked in a learning process can be determined based on a geometric distribution representing a relationship between the length of a frame and a sampling probability.

In one embodiment, the graph representing the first masking scheme (FIG. 5(a)) may have an open sampling probability in which the length of the frame is biased toward a shorter length than a longer length according to a geometric distribution calculated with probability statistics. there is.

For example, when a frame having a relatively short length is masked, the first prediction model predicting the CDR3 sequence of the TCR may record a higher rate of correct answers.

In a further example, when a frame having a relatively short length is masked, the first prediction model predicting the CDR3 sequence of the TCR may record a higher rate of correct answers. These examples exemplarily indicate that prediction of CDR3 sequences of TCRs can be improved in frames of longer length.

In one embodiment, the first prediction model may generate a more improved prediction result for the masking frame of the CDR3 sequence as the number of epochs decreases when frames with a short length are masked in the first training data set.

(b) in FIG. 5 exemplarily shows that the length of a frame may be determined based on an inverse value of a geometric distribution representing a relationship between a length of a frame and a sampling probability.

In one embodiment, the graph representing the second masking method (FIG. 5(b)) is sampling in which the length of the frame is biased to a longer length than a short length according to the inverse value of the geometric distribution calculated with probability statistics. You can have a reformed form of probability.

In one embodiment, it is exemplarily shown that the relationship between the length of the frame and the sampling probability can improve the performance of the first prediction model that tries to predict the CDR3 sequence of a longer TCR based on the inverse of the geometric distribution.

For example, as shown in (b), the inverse value of the geometric distribution calculated with probability statistics may mean that the sampling probabilities for each frame length shown in (a) are allocated in the reverse order.

For example, an inverse value of a geometric distribution calculated through stochastic statistics may mean that a sampling probability of a frame having a relatively long length is assigned greater than a sampling probability of a relatively short length.

(c) in FIG. 5 exemplarily shows that the length of a frame can be determined based on a scheme in which the same sampling probability is applied to the lengths of a plurality of different frames.

As an example, a graph showing the third masking method (FIG. 5(c)) exemplarily shows that the sampling probability is the same regardless of the frame length.

In one embodiment, the 1a prediction model trained through the first masking method (FIG. 5 (a)), the 1b prediction model trained through the second masking method (FIG. 5 (b)), and the third masking method ( At least two or more prediction models among the 1c prediction models trained through FIG. 5(c)) may be operated complementary to each other.

For example, for the training of the CDR3 sequence prediction model of the TCR performed by the computing device 100, the 1c prediction model is driven in the Nth epoch, and the 1a prediction model is driven in the N+1 th epoch, resulting in a predictive model Illustratively shows that they can be operated complementary to each other.

In one embodiment, the first prediction model trained through the first masking method (FIG. 5 (a)), the second masking method (FIG. 5 (b)), and the third masking method (FIG. 5 (c)) is more By generating CDR3 sequence predicted values of various TCRs, it is exemplarily shown that optimal immunogenicity or clinically improved effects can be exhibited.

6 is a first learning data set preprocessed based on data included in guide information generated from HLA and peptides according to an embodiment of the present disclosure, and includes the CDR3 sequence of the TCR through the first prediction model. A method of generating a prediction result is exemplarily shown.

In one embodiment, the second prediction model 620 receives the information of the HLA 612 and the peptide 614 and includes amino acid length information 632, V type 634 and / or J type 636 guide information 630 to be output.

In one embodiment, the first training data set 640 includes genetic information related to the expression of the T cell receptor generated by the second prediction model 620 based on the information of the HLA 612 and the peptide 614 It can be pretreated by V type and J type.

In one embodiment, the first training data set 640 is based on the amino acid length information 632 of the CDR3 sequence of the TCR generated by the second prediction model 620 based on the information of the HLA 612 and the peptide 614. can be pretreated.

In one embodiment, the guide information generated by the second prediction model is the frame position of the CDR3 sequence of the first training data set from the location information where there is a lot of diversity in the V type and/or the location where there is a lot of diversity in the J type. can be determined.

For example, the position of a frame may be used as a meaning encompassing a position of a unit where a frame starts and a position of a unit where a frame ends.

In one embodiment, the frame position of the CDR3 sequence of the first training data set is determined from the location information where the guide information generated by the second prediction model has a lot of diversity in V-type and / or where there is a lot of diversity in J-type can be made Accordingly, the guide information may limit the masking position of the first training data set to achieve learning efficiency of the first prediction model, thereby improving performance of the first prediction model.

In one embodiment, the first training data set may be preprocessed based on information obtained empirically or from a biostatistical database.

In a further embodiment, the predictive model may be trained to output output data corresponding to CDR3 based on input data representing a conjugate of MHC and a peptide.

In one embodiment, MHC and peptides capable of binding or each of MHC and peptides may be obtained through a public database (eg, VDJdb. McPAS. TCR3D, huARdb). In an additional embodiment, bindable MHC and peptides may be determined by a separate artificial intelligence-based model that takes MHC and peptides as inputs and outputs whether bindable or not.

In one embodiment of the present disclosure, a masked learning method may be used in a process of learning a predictive model. After applying a mask to some of the amino acids included in the training data set 640, semi-supervised learning to match the masked amino acids may be performed. The object to which the mask is applied may be limited to amino acid sequences corresponding to the correct answer. Accordingly, the accuracy of the predictive model for predicting CDR3 corresponding to the received peptide and MHC can be further increased.

In a further example, after applying a mask to a frame composed of contiguous units included in the training data set 640, semi-supervised learning to match the units of the masked frame may be performed. Accordingly, the accuracy of the prediction model for predicting the CDR3 of the TCR can be further increased.

In one embodiment of the present disclosure, a plurality of masks corresponding to one frame may be applied to one training data set. In this case, the masked learning method can be implemented by applying X amino acids representing all amino acids or the average of amino acids to another mask in the process of matching one mask among a plurality of masks to be applied to one frame. there is. Accordingly, since a feature called X amino acid is used even in a situation where a plurality of masks exist, prediction accuracy for masked data can be higher.

7 exemplarily illustrates a method of generating a prediction result comprising a CDR3 sequence of a TCR according to an embodiment of the present disclosure. For example, the steps shown in FIG. 7 may be performed by computing device 100 .

In one embodiment, the computing device 100 provides first input data 710a corresponding to MHC, second input data 720a corresponding to peptide, and third input data 730a corresponding to CDR3 of TCR. By performing preprocessing including a grouping process 740 and a segmenting process 750 , training data set 760 may be generated.

In one embodiment, pre-processing may include a grouping process 740 and a segmenting process 750. Grouping process 740 may include grouping the amino acid sequences into N units. The segmenting process 750 may refer to a process of combining grouped groups. The training data set is a data set generated based on the grouping process 740 and the segmenting process 750 and may be used for learning a predictive model.

In one embodiment, the grouping process 740 may include a process of generating units having lengths corresponding to each of the amino acid sequences of different lengths by analyzing the frequency of occurrence for each of the amino acid sequences of different lengths. . Here, the length of the amino acid sequence may correspond to the number of amino acids included in the amino acid sequence. For example, when the length of an amino acid sequence is 4, the number of amino acids included in the amino acid sequence may be 4.

In one example, grouping process 740 can include a uniting process. The unitization process may refer to a process of determining the number of amino acid sequences that can be processed as one unit. A tokenization process may exist as an example of the unitization process. Unitization may be performed on each of the first input data, the second input data, and the third input data based on the units generated by the grouping process 740 .

In one embodiment, the grouping process 740 includes generating a first unit 710b group for a first set of amino acid sequences obtained by analyzing the frequency of occurrence for amino acid sequences having a first length. , And a second set obtained by analyzing the frequency of occurrence of amino acid sequences having a second length shorter than the first length in a state in which the first set of amino acid sequences included in the first group are excluded. It may include generating a group of second units 720b comprising amino acid sequences. Here, units included in the first token group may have a first length, and units included in the second unit 720b group may have a second length.

In one embodiment, the occurrence frequency may represent a value quantitatively representing the probability that a specific amino acid sequence is found within one CDR3 sequence. In one embodiment, the frequency of appearance may represent a value quantitatively representing the ratio of the number of times a specific amino acid sequence is found in all CDR3s to the number of all CDR3s.

In one embodiment, the segmenting process 750 may refer to a process of combining grouped groups. In one example, the segmenting process 750 may include a process of combining tokens to create a training data set. A segment process may include, for example, segment embedding.

In one embodiment, the segmenting process 750 includes first tokens corresponding to 710a and third tokens corresponding to third input data 730a with a separator token interposed therebetween. A second training data set including a first training data set, and second tokens corresponding to the second input data 720a and third tokens corresponding to the third input data 730a with a delimiter token interposed therebetween is generated. can do. Here, the third tokens may be used as correct answer data in the learning process of the predictive model. The predictive model combines the first inference result based on the first training data set and the second inference result based on the second training data set, and corresponds to the first input data 710a corresponding to MHC and the peptide. It can be learned to output the third input data 730a corresponding to CDR3 corresponding to the second input data 720a. Here, the prediction model may include a first prediction model learned based on the first training data set and a second prediction model learned based on the second training data set. In this embodiment, the third input data 730a may be output or learned to output the third input data 730a based on the output of the first prediction model and the output of the second prediction model.

In an embodiment, the prediction result may further include V type and J type of TCR as well as CDR3 corresponding to the first input data 710a and the second input data 720a.

In one embodiment, the predictive model operates based on any type of learning method for generating a prediction result including the third input data 730a from the first input data 710a and the second input data 720a. It can be.

For example, the learning method may include semi-supervised learning in which a mask is applied to amino acids of some of the amino acid sequences and then the masked amino acids are matched.

For example, the learning method may include semi-supervised learning in which a mask is applied to amino acids included in a frame and/or unit of a third sequence corresponding to CDR3 among amino acid sequences, and then matching the masked frame and/or unit. can In this example, the learning method according to an embodiment of the present disclosure applies a mask to a frame and / or unit corresponding to CDR3, and a second sequence corresponding to amino acids and peptides of a first sequence corresponding to MHC A method of not applying a mask may be included for the amino acids of .

In a further embodiment, a learning method for learning a predictive model includes, when the predictive model is learned over a plurality of epochs, applying a mask to a different location on a per-epoch basis or applying a mask of a different size on a per-epoch basis. method can be included. For example, in a first epoch, a mask of a first size may be applied to at least some of amino acids corresponding to CDR3 corresponding to a first position, and in a second epoch different from the first epoch, a second mask different from the first position A mask of a second size may be applied to at least some of the amino acids corresponding to the CDR3 corresponding to the position.

In one embodiment, after the learning of the predictive model 770 is completed, the predictive model 770 receives first input data 710a corresponding to MHC and second input data 710a corresponding to peptide, CDR3 Third input data 730a corresponding to can be output. In an additional embodiment, after learning of the predictive model 770 is completed, the predictive model 770 receives first input data 710a corresponding to MHC and second input data 710a corresponding to peptide, CDR3 , V-type and J-type third input data 730a can be output.

In one embodiment, the prediction model 770 may be trained to output CDR3 and/or V type/J type corresponding to the peptide and MHC based on the training data 760 .

In one embodiment, the first input data 710a may include an amino acid sequence corresponding to MHC. The second input data 710a may include an amino acid sequence corresponding to a peptide. The third input data 730a may include an amino acid sequence corresponding to CDR3, V-type and/or J-type. In the process of learning the predictive model 770, the third input data 730a is CDR3 capable of combining with the first input data 710a corresponding to MHC and the second input data 710a corresponding to peptide (added to this V type and J type) may be included. That is, data on CDR3 corresponding to MHC and peptides may be pre-acquired, and based on the pre-acquired data, learning of the predictive model 770 may be performed.

In one embodiment, the grouping process 740 may generate units 710b, 720b, and 730b corresponding to the input data 710a, 710a, and 730a. In one embodiment, the grouping process 740 generates a first unit 710b corresponding to the first input data 710a, generates a second unit 720b corresponding to the second input data 710a, and , and a third unit 730b corresponding to the third input data 730a may be generated.

In one embodiment, the grouping process 740 includes a process for dividing the input data into smaller units or sizes, where the divided units may correspond to tokens.

In one embodiment, the segmenting process 750 may include a process of creating a training data set 760 based on tokens.

The segmenting process 750 according to an embodiment of the present disclosure generates a training data set 760 that includes first input data 710a corresponding to MHC and third input data 730a corresponding to CDR3. can do. In this learning data set 760, the third input data 730a corresponding to the first input data 710a may be used as correct answer data. The segmenting process 750 may generate a training data set 760 including second input data 710a corresponding to a peptide and third input data 730a corresponding to CDR3. In the learning data set 760, the third input data 730a corresponding to the second input data 710a may be used as correct answer data. In this embodiment, the segmenting process 750 generates two types of training data sets 760, and each of the generated two types of training data sets 760 is independently applied to the predictive model 770. can In this case, the prediction model 770 may be a single integrated model or may be composed of a plurality of separate models for receiving and processing each training data set 770 .

The segmenting process 750 according to an embodiment of the present disclosure includes first input data 710a corresponding to MHC, second input data 710a corresponding to peptide, and third input data 730a corresponding to CDR3. A training data set 760 including ) can be created. In this learning data set 760, the third input data 730a corresponding to the first input data 710a and the second input data 710a may be used as correct answer data. In this embodiment, the segmenting process 750 may generate input data in a concatenated form of the first input data 710a and the second input data 710a, and the third input data 730a as the above. The training data set 760 may be created in a manner of using it as correct answer data corresponding to the input data. In this case, through the segmenting process 750, for example, input data in an integrated form may be generated in such a way that amino acid sequences corresponding to MHC and amino acid sequences corresponding to peptides are concatenated with each other.

In an embodiment, concatenated first and second input data may be used in the inference process of the predictive model 770, and CDR3 corresponding to the concatenated input data may be output. In an embodiment, in the inference process of the predictive model 770, the first input data and the second input data are obtained by combining CDR3 for the first input data and CDR3 for the second input data or inputting them to another model. Corresponding CDR3 can be output.

8 is a schematic diagram of a computing environment according to one embodiment of the present disclosure.

A component, module or unit in this disclosure includes routines, procedures, programs, components, data structures, etc. that perform particular tasks or implement particular abstract data types. In addition, those skilled in the art will understand that the methods presented in this disclosure can be used in single-processor or multiprocessor computing devices, minicomputers, mainframe computers as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and the like ( It will be fully appreciated that each of these may be implemented with other computer system configurations, including those that may be operative in connection with one or more associated devices.

Embodiments described in this disclosure may also be practiced in distributed computing environments where certain tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote memory storage devices.

A computing device typically includes a variety of computer readable media. Computer readable media can be any medium that can be accessed by a computer, including volatile and nonvolatile media, transitory and non-transitory media, removable and non-transitory media. Includes removable media. By way of example, and not limitation, computer readable media may include computer readable storage media and computer readable transmission media.

Computer readable storage media are volatile and nonvolatile media, transitory and non-transitory, 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. includes media Computer readable storage media may include RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital video disk (DVD) or other optical disk storage device, magnetic cassette, magnetic tape, magnetic disk storage device or other magnetic storage device. device, or any other medium that can be accessed by a computer and used to store desired information.

A computer readable transmission medium typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism. Including all information delivery media. The term modulated data signal means a signal that has one or more of its characteristics set or changed so as to encode information within the signal. By way of example, and not limitation, computer readable transmission media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, and other wireless media. Combinations of any of the above are also intended to be included within the scope of computer readable transmission media.

An exemplary environment 2000 implementing various aspects of the present invention is shown comprising a computer 2002, which includes a processing unit 2004, a system memory 2006 and a system bus 2008. do. Computer 200 herein may be used interchangeably with a computing device. System bus 2008 couples system components, including but not limited to system memory 2006, to processing unit 2004. Processing unit 2004 may be any of a variety of commercially available processors. Dual processors and other multiprocessor architectures may also be used as the processing unit 2004.

System bus 2008 may be any of several types of bus structures that may additionally be interconnected to a memory bus, a peripheral bus, and a local bus using any of a variety of commercial bus architectures. System memory 2006 includes read only memory (ROM) 2010 and random access memory (RAM) 2012 . The basic input/output system (BIOS) is stored in non-volatile memory 2010, such as ROM, EPROM, EEPROM, etc. BIOS is a basic set of information that helps transfer information between components within the computer 2002, such as during startup. contains routines. RAM 2012 may also include high-speed RAM, such as static RAM, for caching data.

The computer 2002 may also read from an internal hard disk drive (HDD) 2014 (eg EIDE, SATA), a magnetic floppy disk drive (FDD) 2016 (eg a removable diskette 2018), or for writing to them), SSDs and optical disk drives 2020 (for example, for reading CD-ROM disks 2022 or reading from or writing to other high capacity optical media such as DVDs) include The hard disk drive 2014, magnetic disk drive 2016, and optical disk drive 2020 are connected to the system bus 2008 by the hard disk drive interface 2024, magnetic disk drive interface 2026, and optical drive interface 2028, respectively. ) can be connected to The interface 2024 for external drive implementation includes, for example, at least one or both of Universal Serial Bus (USB) and IEEE 1394 interface technologies.

These drives and their associated computer readable media provide non-volatile storage of data, data structures, computer executable instructions, and the like. In the case of computer 2002, drives and media correspond to storing any data in a suitable digital format. Although the description of computer-readable storage media above refers to HDDs, removable magnetic disks, and removable optical media such as CDs or DVDs, those skilled in the art can use zip drives, magnetic cassettes, flash memory cards, cartridges, It will be appreciated that other types of computer-readable storage media, such as those of other types, may also be used in the exemplary operating environment and that any such media may contain computer-executable instructions for performing the methods of the present invention. .

A number of program modules may be stored on the drive and RAM 2012, including an operating system 2030, one or more application programs 2032, other program modules 2034, and program data 2036. All or portions of the operating system, applications, modules and/or data may also be cached in RAM 2012. It will be appreciated that the present invention may be implemented in a variety of commercially available operating systems or combinations of operating systems.

A user may enter commands and information into the computer 2002 through one or more wired/wireless input devices, such as a keyboard 2038 and a pointing device such as a mouse 2040. Other input devices (not shown) may include a microphone, IR remote control, joystick, game pad, stylus pen, touch screen, and the like. Although these and other input devices are often connected to the processing unit 2004 through an input device interface 2042 that is connected to the system bus 2008, a parallel port, IEEE 1394 serial port, game port, USB port, IR interface, may be connected by other interfaces such as the like.

A monitor 2044 or other type of display device is also connected to the system bus 2008 through an interface such as a video adapter 2046. In addition to the monitor 2044, computers typically include other peripheral output devices (not shown) such as speakers, printers, and the like.

Computer 2002 may operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 2048 via wired and/or wireless communications. Remote computer(s) 2048 may be workstations, server computers, routers, personal computers, handheld computers, microprocessor-based entertainment devices, peer devices, or other common network nodes, and generally relate to computer 2002. Although many or all of the described components are included, for brevity, only memory storage device 2050 is shown. The logical connections shown include wired/wireless connections to a local area network (LAN) 2052 and/or a larger network, such as a wide area network (WAN) 2054 . Such LAN and WAN networking environments are common in offices and corporations and facilitate enterprise-wide computer networks, such as intranets, all of which can be connected to worldwide computer networks, such as the Internet.

When used in a LAN networking environment, computer 2002 is connected to local network 2052 through wired and/or wireless communication network interfaces or adapters 2056. Adapter 2056 may facilitate wired or wireless communications to LAN 2052, which also includes a wireless access point installed therein to communicate with wireless adapter 2056. When used in a WAN networking environment, computer 2002 may include a modem 2058, be connected to a communications server on WAN 2054, or other means of establishing communications over WAN 2054, such as over the Internet. have Modem 2058, which can be internal or external and wired or wireless, is connected to system bus 2008 through serial port interface 2042. In a networked environment, program modules described for computer 2002, or portions thereof, may be stored in remote memory/storage device 2050. It will be appreciated that the network connections shown are exemplary and other means of establishing a communication link between computers may be used.

Computer 1602 is any wireless device or entity that is deployed and operating in wireless communication, such as printers, scanners, desktop and/or portable computers, portable data assistants (PDAs), communication satellites, and associated with wireless detectable tags. It operates to communicate with arbitrary equipment or places and telephones. This includes at least Wi-Fi and Bluetooth wireless technologies. Thus, the communication may be a predefined structure as in conventional networks or simply an ad hoc communication between at least two devices.

It is to be understood that the specific order or hierarchy of steps in the processes presented is an example of exemplary approaches. Based upon design priorities, it is to be understood that the specific order or hierarchy of steps in the processes may be rearranged within the scope of the present disclosure. The method claims of this disclosure present elements of the various steps in a sample order, but are not meant to be limited to the specific order or hierarchy presented.

Claims (20)

As a method of training an artificial intelligence-based predictive model performed by a computing device,
Acquiring a first training data set related to CDR3 of a T Cell Receptor (TCR), wherein the first training data set includes guide information to be used in learning a first artificial intelligence-based prediction model;
Based on the guide information, pre-processing the first training data set for learning of the first predictive model - the pre-processing includes a predetermined number of consecutive units included in the first training data set ( including applying a mask to a frame containing units; and
Using the preprocessed first learning data set, the first prediction model predicts the amino acid sequence of the CDR3 of the TCR corresponding to the major histocompatibility complex (MHC) and the peptide, the first prediction model learning step;
Including,
The length of the frame corresponds to the number of consecutive units included in the frame,
method. delete According to claim 1,
The length of the frame is
Determined based on a geometric distribution representing the relationship between the length of the frame and the sampling probability,
method. According to claim 1,
The length of the frame is
Determined based on the inverse value of the geometric distribution representing the relationship between the length of the frame and the sampling probability,
method. According to claim 1,
The length of the frame is
Determined based on how the same sampling probability is applied for the length of a plurality of different frames,
method. According to claim 1,
The pre-processing step is:
determining a position of a frame including a predetermined number of consecutive units within the first training data set based on the guide information; and
applying a mask to the frame based on the determined position of the frame;
including,
method. According to claim 1,
The pre-processing step is:
determining a position of a frame including a predetermined number of consecutive units in the first training data set based on the V type and the J type included in the guide information; and
applying a mask to the frame based on the determined position of the frame;
including,
method. According to claim 1,
determining a position of a frame including a predetermined number of contiguous units in the first training data set, based on the V type, J type, and length of the amino acid sequence included in the guide information; and
applying a mask to the frame based on the determined position of the frame;
including,
method. According to claim 1,
The guide information is obtained from the output of a second prediction model based on artificial intelligence different from the first prediction model, and
The second predictive model outputs a prediction result including the guide information based on the peptide and MHC,
method. According to claim 1,
The step of learning the first predictive model,
Within a frame composed of a predetermined number of contiguous units, each of the units other than the masked single unit is substituted with an amino acid X and the masked single unit is converted into the first prediction model by using the substituted amino acid X training the first predictive model to make the prediction;
including,
method. According to claim 10,
Location information including a distance to the masked single unit is assigned to each of units other than the masked single unit in the frame, and
The step of learning the first predictive model,
training the first prediction model so that the first prediction model predicts the masked single unit based on the location information;
including,
method.
According to claim 1,
The step of learning the first predictive model,
a first loss function determined based on a difference between an output representation of a unit included in the preprocessed first training data set and a correct answer of the unit;
Determined based on the output representation of units before the start position of the frame to which the mask is applied in the first training data set, the output representation of units after the end position of the frame, and the embedding position of each unit in the frame a second loss function that becomes;
Output expression of the masked single unit according to a method of substituting each of the units other than the single unit masked with amino acid X in a frame composed of a predetermined number of consecutive units in the preprocessed first training data set a third loss function determined based on a difference between ? and the correct answer of the masked single unit; or
a fourth loss function determined based on a difference between the CDR3 sequence output through the first prediction model and the correct answer of the CDR3 sequence of the TCR corresponding to the actual peptide and MHC;
training the first predictive model based on at least two of;
including,
method. According to claim 12,
The second loss function is assigned a higher weight than the first loss function,
method. According to claim 1,
Training the first predictive model to:
Determining a CDR3 sequence that cannot be output from the peptide and MHC from the subject from the information of the peptide, MHC and corresponding TCR CDR3 from the subject; and
learning the first prediction model by excluding the determined non-outputable CDR3 sequence from the first training data set;
including,
method. 15. The method of claim 14,
The CDR3 sequence of the TCR to be excluded from the first learning data set is determined based on data included in the biostatistical database,
method. 15. The method of claim 14,
The CDR3 sequence of the TCR to be excluded from the first learning data set is determined based on data obtained by sequencing the subject by molecular biological experimental methods.
method. According to claim 1,
Applying a mask to the frame,
When the first prediction model predicts a mask corresponding to each of the consecutive units in the frame, applying an amino acid X reflecting factors of amino acid species to remaining masks other than the mask to be predicted;
including,
method. According to claim 1,
The predictive model is learned by applying a mask to frames containing a predetermined number of consecutive units, and
The prediction model is based on a start value corresponding to a unit before the start position of a frame to which the mask is applied and an end value corresponding to a unit after the end position of the frame in the first training data set, the mask is applied Learning by matching the value corresponding to at least one unit included in the frame to be,
method. A computer program stored on a computer-readable storage medium, which, when executed by a computing device, causes the computing device to perform operations for learning an artificial intelligence-based predictive model, the operations comprising:
Acquiring a first training data set related to CDR3 of a T Cell Receptor (TCR), wherein the first training data set includes guide information to be used in learning a first artificial intelligence-based prediction model;
Based on the guide information, pre-processing the first training data set for learning of the first predictive model - the pre-processing operation includes a predetermined number of consecutive units included in the first training data set ( Includes an operation of applying a mask to a frame including units; And using the preprocessed first learning data set, the first predictive model predicts the amino acid sequence of the CDR3 of the TCR corresponding to the major histocompatibility complex (MHC) and the peptide. an operation of learning;
Including,
The length of the frame corresponds to the number of consecutive units included in the frame,
A computer program stored on a computer readable storage medium. As a computing device,
at least one processor; and
Memory;
Including,
The at least one processor is:
Acquiring a first training data set related to CDR3 of a T Cell Receptor (TCR), wherein the first training data set includes guide information to be used in learning of a first artificial intelligence-based prediction model;
Based on the guide information, pre-processing the first training data set for learning the first predictive model - the pre-processing operation includes a predetermined number of consecutive units included in the first training data set ( Includes an operation of applying a mask to a frame including units; And using the preprocessed first learning data set, the first predictive model predicts the amino acid sequence of the CDR3 of the TCR corresponding to the major histocompatibility complex (MHC) and the peptide. an operation of learning;
to perform,
The length of the frame corresponds to the number of consecutive units included in the frame,
computing device.

Images