KR102171681B1 - Computer readable media recording program of consructing potential rna aptamers bining to target protein using machine learning algorithms and process of constructing potential rna aptamers - Google Patents

Abstract

The present invention is a computer-readable recording medium recording a program for generating and verifying a candidate RNA aptamer sequence that binds to a target protein molecule through a random forest model, which is one of machine learning models, and binds to a target protein molecule using the same. It relates to a method for generating and verifying a candidate RNA aptamer sequence. By using the program according to the present invention, the pool of candidate aptamer sequences that bind to a specific target protein is reduced, and the final aptamer can be selected quickly and efficiently.

Description

A computer-readable recording medium recording a program that creates a candidate RNA aptamer that binds to a target protein using a machine learning algorithm, and a method for generating a candidate RNA aptamer {COMPUTER READABLE MEDIA RECORDING PROGRAM OF CONSRUCTING POTENTIAL RNA APTAMERS BINING TO TARGET PROTEIN USING MACHINE LEARNING ALGORITHMS AND PROCESS OF CONSTRUCTING POTENTIAL RNA APTAMERS}

This patent application is part of the “Personal Basic Research” research project of the Ministry of Science, Technology and Communication of the Republic of Korea, and “Development of a predictive model for protein-nucleic acid interaction” (Host: Inha University; Project number: 2015R1A1A3A04001243) Cancer gene discovery through mathematical modeling of big data and individual gene network inference" (Host: Inha University; Assignment number: 2017R1E1A1A03069921) This is the result of the project.

The present invention relates to a medium in which a program for generating a molecule that binds to a target substance is recorded, and more particularly, a program for generating and constructing a candidate RNA aptamer that binds to a target protein through a learning process using a machine learning algorithm. It relates to a recording medium that can be read by a recorded computer, and a method of generating and constructing a candidate RNA aptamer that binds to a target protein by using the recording medium.

As the understanding of molecules related to biological mechanisms deepens, interest in interactions between biological molecules is increasing. Among these biological molecules, aptamers have recently attracted attention. An aptamer is a single-stranded nucleic acid that has a stable three-dimensional structure by itself and can bind with high affinity and specificity for a target molecule. . The aptamer binding to the target molecule is similar to a monoclonal antibody, but has the following advantages compared to a monoclonal antibody.

Antibodies have a large molecular structure of approximately 150 kDa and are difficult to manufacture and modify, but aptamers, which are generally composed of 100 or less nucleotide sequences, have a small molecular structure and are therefore easily modified. Since the aptamer has very good stability compared to the antibody, it can be transported and/or stored at room temperature and can maintain its function even after sterilization. In addition, since the aptamer can be regenerated in a short time even when denaturation occurs, it can be easily applied as a diagnostic material requiring long or repeated use.

Moreover, since antibodies are manufactured using animals or cells, a lot of time and cost are required, and there is a possibility that the functionality may vary depending on the timing or method of manufacture (batch to batch variation). However, since the aptamer is manufactured using a chemical synthesis method, it can be manufactured in a short time and at low cost, there is little batch to batch variation, and a high purity purification process is very easy. In addition, in the case of antibodies or other pharmaceutical proteins, immune rejection reactions frequently occur in vivo, but aptamers are known to hardly cause immune rejection reactions in vivo, which is advantageous in the development of therapeutic materials. In addition, aptamers can be prepared for toxins, complex protein complexes, or sugar-protein complexes that are difficult to make antibodies, and the flexibility to bind to new target substances makes it a novel aptamer. It can be actively used for excavation.

Regarding the screening of new aptamers, a technique called SELEX (Systematic Evolution of Ligands by EXponential enrichment), developed in 1990 by Larry Gold's research team at the University of Colorado, is basically used. Looking at the process of discovering new aptamers through SELEX, 1) using DNA synthesis and in vitro transcription ( in case of RNA) to prepare various types of nucleic acid libraries (>10 ⁵ ), and 2) nucleic acid structures Nucleic acid constructs (aptamer candidate molecules) in the library are screened only for nucleic acid constructs capable of binding to a desired target molecule. In the selection process, nucleic acid constructs that are not bound to the target molecule are washed through a method such as affinity chromatography, and only binding to the target molecule is selectively obtained. Finally, the nucleic acid construct is separated from the target molecule (elution), and the separated nucleic acid is amplified by a method such as PCR (polymerase chain reaction). Subsequently, the selection and separation process using only the amplified nucleic acid construct is repeated 5 to 15 times, thereby discovering an aptamer showing very excellent binding power and specificity.

However, in order to select a candidate aptamer molecule, a random nucleic acid library of 10 ¹⁵ or more is usually prepared, and only a small portion of the nucleic acid sequence is bound to the target molecule. When a lot of time, labor and cost are required. In addition, due to the variety of aptamers in vivo antigen density, interaction and microenvironment (microenvironment) from the (in vivo) identified by SELEX it can not be combined with the target molecule in vivo.

Regarding the problem of selecting an aptamer, U.S. Patent No. 9,315,804 proposes a method of identifying at least one aptamer for a target substance from a candidate aptamer in consideration of a fitness function. However, even in the US patent, since there is still a large pool of initial aptamer candidate substances, a considerable amount of time and labor must be invested when selecting an aptamer that strongly binds to the target substance.

The present invention has been proposed in order to solve the problems of the prior art, and an object of the present invention is to be read with a computer recording a program for generating and constructing a potential candidate aptamer that binds to a target substance in a fast and efficient manner. It is intended to provide a method for generating and constructing a potential candidate aptamer that binds to a possible recording medium and a target material.

According to one aspect of the present invention, the present invention constructs a feature vector of an RNA sequence and a feature vector of an amino acid sequence constituting a protein based on the RNA-protein complex data, and a random forest based on the constructed feature vector. Sequence learning means for training RNA sequence and protein sequence by applying the model; And a sequence discovery means for filtering a random RNA sequence and applying a random forest model learned through the sequence learning means to construct a candidate RNA aptamer that binds to a target protein molecule among the filtered RNA sequences. , In the sequence learning means, the RNA feature vector has an interaction propensity with amino acids constituting a protein, a mono-nucleotide (mono-nucleotide; single-base) composition of the RNA sequence, and a di-nucleotide; It is written based on the composition of 2-base) and pseudo tri-nucleotide (pseudo tri-nucleotide), and in the sequence learning means, the protein feature vector determines the composition-transition-distribution and similar amino acid composition of amino acids constituting the protein. It provides a computer-readable recording medium in which a program for generating a candidate RNA aptamer that binds to a target protein molecule to be prepared is recorded.

In an alternative embodiment, the program may further include a sequence evaluation means for evaluating a model for generating the candidate RNA aptamer selected by the sequence discovery means using an evaluation scale.

For example, the evaluation scale is Sensitivity, Specificity, Accuracy, Positive predictive value, Negative predictive value, and Matthews correlation coefficient. It is characterized in that at least one selected from.

In one example, the sequence discovery means filters the random RNA sequence by limiting the secondary structure, free energy, the number of unpaired nucleotides, and the number of identical secondary structures in the pool. Characterized in that.

Meanwhile, the similar tri-nucleotide composition constituting the RNA feature vector includes hydrophobicity, hydrophilicity, and side-chain mass of the tri-nucleotide, and constitutes the protein feature vector. The similar amino acid composition may include hydrophobicity, hydrophilicity, side chain-weight, ionization index, and isoelectric point of the amino acid.

According to another aspect of the present invention, the present invention is a method for generating a candidate RNA aptamer that binds to a target protein molecule using a computer-implemented program, by means of sequence learning, based on RNA-protein complex data. Constructing the feature vector of the RNA sequence and the feature vector of the amino acid sequence constituting the protein, and training the RNA sequence and the protein sequence by applying a random forest model based on the constructed feature vector; And filtering the random RNA sequence by the sequence discovery means, and applying the random forest model learned through the sequence learning means, to construct a candidate RNA aptamer that binds to the target protein molecule among the filtered RNA sequences. Including a step, wherein in the sequence learning means, the RNA feature vector has an interaction propensity with amino acids constituting a protein, a mono-nucleotide (single-base) composition of the RNA sequence, a di-nucleotide ( It is written based on the composition of di-nucleotide (2-base) and pseudo tri-nucleotide (pseudo tri-nucleotide), and in the sequence learning means, the protein feature vector is the composition-transition-distribution of amino acids constituting the protein. And a method of generating a candidate RNA aptamer that binds to a target protein molecule prepared based on a similar amino acid composition.

If necessary, it may further include the step of evaluating the model for generating the candidate RNA aptamer constructed by the sequence discovery means by the sequence evaluation means using an evaluation scale.

According to the program and method of the present invention, as a machine learning algorithm, training using a random forest algorithm that improves the shortcomings of a decision tree, and using a learning model to generate an aptamer with a very high possibility of binding to a specific target molecule, Can build. An initial pool is formed only with potential aptamer candidates with a high possibility of binding to a target molecule, and the final aptamer can be quickly and efficiently selected starting from the nucleic acid library in the reduced pool.

Since the final aptamer can be quickly and efficiently screened only with potential aptamer candidates with a high possibility of binding to a specific target molecule, new drug candidates that can competitively bind to the aptamer for the new drug target protein are selected. Aptamer or aptamer to measure the concentration of environmentally hazardous substances such as pollutants or food hazardous substances, development of diagnostic reagents or microarrays that use aptamers as biomarkers, and development of new drug candidate substances -It is expected that the present invention can be utilized when developing a biosensor that detects a signal according to whether or not a target molecule is bound.

1 is a computer equipped with a program for constructing and generating a candidate RNA aptamer that binds to a target protein molecule using a random forest algorithm, which is one of machine learning algorithms, according to an exemplary embodiment of the present invention, and the program is recorded. It is a diagram schematically showing the configuration of a computer-readable recording medium.
2 is a diagram schematically showing an entire framework for discovering candidate RNA aptamer sequences according to an exemplary embodiment of the present invention. mC is a mono-nucleotide (single-base) composition, dC is a di-nucleotide (2-base) composition, and PseTNC is a pseudo tri-nucleotide composition (pseudo tri-nucleotide) Composition), CTD represents composition-transition-distribution of amino acid groups, and PseAAC represents pseudo amino acid composition.
3 is a flow chart schematically showing a method of constructing and generating a candidate RNA aptamer that binds to a target protein molecule using a random forest algorithm according to an exemplary embodiment of the present invention.
4 shows an example of a positive window and a negative window of an RNA sequence consisting of 27 nucleotides in an RNA sequence, according to an exemplary embodiment of the present invention. As the RNA sequence is scanned using a sliding window of 27 nucleotides, a feature vector representing the window is generated. + Represents a nucleotide that binds to a protein, and-represents a nucleotide that does not bind to a protein. If the intermediate nucleotide in a window is a nucleotide that binds to a protein, the feature vector encoding that window is considered positive. On the other hand, if one window contains only nucleotides that do not bind to the protein, the feature vector for that window is considered to be negative. Feature vectors that are neither positive nor negative are not used in the learning process.
5 shows an example of a secondary structure of a known RNA aptamer. The secondary structure has a structure that is closed by at least three base pairs, with or without dangling end bases, and the nucleotide that best binds to the target protein molecule indicated by a red circle. Is observed in the single-stranded portion of the corresponding RNA secondary structure.

Hereinafter, the present invention will be described with reference to the accompanying drawings when necessary.

Interactions between nucleic acid molecules and protein molecules are essential in various cellular processes. Recently, high-throughput technology has generated a lot of data on the interaction between protein molecules and nucleic acid molecules, so that binding sites in a specific sequence or interactions between these sequences are predicted. In order to decide, it is necessary to develop a method using a computer model.

In particular, a computer model capable of discovering and generating a novel aptamer sequence for a target protein molecule, beyond a simple classification model that simply determines whether a nucleic acid sequence binds to a target protein molecule or applied only to a specific target protein molecule. Need to develop. According to the present invention, a machine learning model called Random Forest (RF) is used to generate, select, evaluate and/or verify candidate RNA aptamer sequences that bind to a target protein molecule. Explain.

1 is a computer equipped with a program for constructing and generating a candidate RNA aptamer binding to a target protein molecule using a random forest algorithm/model according to an exemplary embodiment of the present invention, and a computer readable program on which the program is recorded. A diagram schematically showing the configuration of a recording medium, and FIG. 2 schematically illustrates a method of constructing and generating a candidate RNA aptamer binding to a target protein molecule by applying a random forest model according to an exemplary embodiment of the present invention. It is a flow chart shown.

As shown in FIG. 1, a medium 200 in which a program 210 for generating a candidate RNA aptamer sequence binding to a target protein molecule is recorded may be mounted on an appropriate computer 100. The computer 100 is a data processing device having a minimum data processing capability.

For example, the computer 100 may be a computer terminal such as a desktop or a notebook computer and/or a mobile terminal such as a smartphone, a tablet PC, or a PDA. If necessary, the computer 100 may be connected to the Internet or a data communication network such as 3G, LTE and/or 5G. For example, the computer 100 may be connected to a server (not shown) through a communication network, or may be connected to the server in a local form, but the computer according to the present invention is not necessarily connected to the communication network.

The computer 100 may include various hardware 110 to 140 and 160 and various components that are software such as driving/application programs 150 for executing such hardware.

In an exemplary embodiment, the computer 100 trains a suitable candidate aptamer using a program that generates data or information, for example, a candidate aptamer, which is a kind of application program, such as a keyboard, mouse, and/or touch panel, It has an input unit 110 capable of creating an initial input value for learning. The computer 100 also includes an output unit 120, such as a printer, for printing input or downloaded data or information, and a display 130, such as a monitor, for displaying these data on a screen. In addition, the computer 100 includes various data and information (eg, the amino acid sequence of the target protein molecule or the secondary structure data of the candidate RNA aptamer; the generated candidate aptamer sequence; a known conventional RNA aptamer sequence and / Or a secondary structure, etc.), such as RAM, ROM, hard disk, and the like, and a driving/application program 150 is mounted for various implementation tasks in the computer 100.

As an example of the driving/application program 150 implemented in the computer 100, not only driving software such as WINDOWS, but also content editing software capable of creating and editing contents such as documents, images, and video data, and a communication network such as the Internet Browsers for retrieving data or information from the network, training using artificial neural networks, and software for building learning models. However, in addition to that, various driving/application programs may be mounted on the computer 100.

Further, the computer 100 includes a control unit 160 such as a CPU so as to properly control the operations of these hardware and software. In particular, the memory 140 and the control unit 160 in the computer 100 constitute a kind of program execution unit in that they realize a series of operations according to the instructions of the program 210 recorded in the recording medium 200.

Meanwhile, the recording medium 200 that can be mounted on the computer 100 so that the computer 100 can read it is a program 210 that generates, selects, and verifies a candidate aptamer sequence that binds to a target protein molecule according to the present invention. ) Is recorded. As described later, a candidate RNA aptamer sequence that binds to a target protein molecule can be efficiently generated by using the program 210 according to the present invention. For example, the computer-readable recording medium 200 according to the present invention includes a series of program instructions for generating, selecting, evaluating, and verifying a candidate aptamer sequence that binds to a target protein molecule, as well as data files and data. Structures and the like may be included alone or in combination.

An example of a computer-readable recording medium 200 on which a program 210 for generating a candidate aptamer sequence that binds to a candidate protein sequence according to the present invention is recorded is all types of data that can be read by a computer system. Includes a recording device. Examples of the recording medium 200 readable by the computer 100 include magnetic media such as magnetic tapes, hard disks, and floppy disks, optical recording media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -An optical medium and a hardware device specially configured to store and execute program instructions such as ROM, RAM, flash memory, and the like.

In addition, a carrier wave (for example, transmitted through the Internet) is also included. The recording medium 200 includes an optical or metal wire including a carrier wave that transmits a signal specifying a program command, a data structure, etc. , May be a transmission medium such as a waveguide. The recording medium 200 readable by the computer 100 may be distributed over a computer system connected through a communication network, and codes readable by the computer 100 may be stored and executed in a distributed manner. On the other hand, examples of the program instruction include not only machine language codes produced by a compiler but also high-level language codes that can be executed by a computer using an interpreter or the like. Functional programs, codes and code segments for implementing the present invention, which are exemplarily described below, may be easily deduced by ordinary programmers in the art to which the present invention pertains.

A program that generates a candidate RNA aptamer that binds to a target protein molecule recorded in the recording medium 200 (hereinafter, abbreviated as an RNA aptamer generation program, 210) is an RNA sequence and protein amino acid sequence through appropriate learning and training models. Sequence learning means 220 for training, and sequence discovery means 230 for constructing an appropriate candidate RNA aptamer binding to the target protein molecule from the randomly generated RNA sequence, and, if necessary, sequence discovery means 230 It may include a sequence evaluation means 240 for evaluating the suitability of the candidate RNA aptamer constructed in.

A process of discovering a candidate RNA aptamer according to the present invention will be described with reference to FIG. 3. 3 is a diagram schematically showing an entire framework for discovering a candidate RNA aptamer that binds to a target molecule according to an exemplary embodiment of the present invention.

Sequence learning means 220 constructs an RNA feature vector and a protein feature vector using a known database having protein-RNA complex data, and a random forest (RF) model from the constructed feature vector To train RNA and protein sequences.

In one exemplary embodiment, an in vivo analysis method such as CLIPdb that provides a protein-RNA complex data set analyzed and identified according to cross-linking and immunoprecipitation (CLIP; clip) It is possible to utilize a known database that provides a data set of protein-RNA complexes identified through clips and their modifications and improved analysis methods.

For example, the in vivo analysis method, which is a clip and a modification thereof, is 1) CLIP-seq (CLIP sequencing) that combines UV cross-linking and immunoprecipitation to identify the binding site of RNA-protein. ; HITS (high-throughput sequencing)-CLIP), 2) PAR-CLIP (photoactivatable ribonucloside-enhanced cross-linking and immunoprecipitation), 3) protein used to identify ribonucleoprotein complexes containing RNA binding proteins or microRNAs ICLIP (individual nucleotide-resolution cross-linking and immunoprecipitation), which uses UV light to covalently bind RNA molecules, 4) sCLIP (simple CLIP), which reduces the amount of RNA and omits radiolabeling of immunoprecipitated RNA. have.

In another exemplary embodiment, the in vitro analysis method SELEX (Systematic Evolution of Ligands by EXponential enrichment) and its modifications, modifications, and other protein-RNA complexes providing a data set identified through the in vitro analysis method. You can use a known database.

For example, SELEX is a protein identified by repeating the process of 1) preparing a nucleic acid library, 2) selecting a nucleic acid structure that binds to a target protein molecule using affinity chromatography, etc., and 3) separating and amplifying the nucleic acid structure. Data sets for RNA complexes can be provided.

On the other hand, a protein-RNA complex data set identified through other methods of improving and modifying SELEX can be utilized to learn the sequence. Other methods of improving and modifying SELEX are: 1) Spiegelmer method using L-oligo-nucleotide, which is not degraded by nuclease, and 2) binding with proteins present on the cell surface without high-purity protein purification. Cell SELEX, which discovers specific aptamers in a cell-to-Aptamer method, 3) capillary electrophoresis SELEX (CE-SELEX) using capillary electrophoresis, 4) counter-SELEX, 5) Toggle SELEX, etc. . In addition, in order to improve the initial aptamer obtained through SELEX into a stable and powerful aptamer, 1) Ribose 2'-OH of RNA aptamer is 2'-F, 2'-NH ₂ , 2'-O-methyl Group substitution, or a post-SELEX process of conjugating aptamer to a polymer such as polyethylene glycol (PEG), diacylglycerol or cholesterol may be performed.

The sequence learning means 220 constructs an appropriate feature vector for these sequences in order to train and learn the RNA sequence and the amino acid sequence of the protein. In the present invention, the characteristic vector for the RNA sequence is the interaction propensity (IP) with the amino acid of the RNA, the mono-nucleotide composition of the RNA sequence (mC, single-base composition), the di-nucleotide It can be prepared based on the composition (di-nucleotide composition; dC, 2-base composition) and pseudo tri-nucleotide composition (PseTNC, pseudo 3-base composition).

When predicting RNA-protein interactions, the tendency of interaction of nucleotide triplets with amino acids (IP), which is one of the basic characteristics of RNA feature vectors, can be a very strong feature. Other features can also be utilized as features of the RNA sequence.

On the other hand, a feature vector for training the amino acid sequence of a protein can be created based on the composition-transition-distribution (CTD) and pseudo amino acid composition (PseAAC) of amino acids constituting the protein. have. In order to extract protein features, the 20 amino acids constituting the protein can be clustered according to appropriate criteria. For example, amino acids constituting a protein may be clustered based on a dipole and a volume, but the present invention is not limited thereto.

In addition, with respect to the pseudo 3-base composition (PseTNC), which is one of the features for training the RNA sequence, and the pseudo-amino acid composition (PseAAC), which is one of the features for training the amino acid sequence of a protein, the nucleotides constituting the RNA sequence Physicochemical properties can be extracted. For example, the hydrophobicity, hydrophilicity and side-chain mass of the nucleotides constituting the nucleotide in the protein-RNA complex known in relation to the similar 3-base composition It can be considered as a property of this nucleotide. On the other hand, with regard to the composition of similar amino acids, the hydrophobicity, hydrophilicity, side chain-weight, ionization index (e.g., the ionization index of the carboxy terminal and the amino group terminal), and the isoelectric point of the amino acids constituting the protein in the known protein-RNA complex should be considered. I can. However, in addition to that, other features of the nucleotides constituting RNA and/or amino acids constituting the protein may be utilized as needed.

When calculating these pseudo 3-base composition and similar amino acid composition, a weight factor and a counted rank of the correlation formed along the amino acid sequence of the protein (tier, usually expressed as λ) are set. I can. For example, as the weight factor, an arbitrary value between 0.05 and 0.70 may be set. Meanwhile, the correlation tier must be 1) smaller than the length of the input protein sequence, 2) must not be a negative integer, and 3) when the tier value is 0, the output of the similar amino acid composition (PseAAC) is a normal amino acid. It needs to be set to be designated by composition. According to one exemplary embodiment, the correlation tier value of the amino acid sequence constituting the protein may be set to an integer between 1 and 5.

On the other hand, there are many cases where there are 30 or less RNA aptamers finally identified in experimental techniques such as SELEX. In addition, in the case of the PAR-CLIP technique, a protein-binding RNA sequence typically composed of 21 to 35 nucleotides can be identified. Therefore, it may be desirable to design an RNA sequence having a length corresponding to a nucleotide identified through this analysis method as a sequence for training and learning. Accordingly, the number of nucleotides in the RNA sequence to be trained may be set to 20 to 40, preferably 21 to 35, but the present invention is not limited thereto.

Based on the known protein-RNA complex data set, when a feature vector of the RNA sequence and the amino acid sequence of the protein is constructed, the sequence learning means 230 uses a machine learning algorithm (or model) to Train sequence and amino acid sequence. For example, the sequence learning means 230 may train a feature vector of an RNA sequence and an amino acid sequence of a protein using a random forest (RF) model. The random forest model is one of the algorithms for improving the shortcomings of the decision tree, and is known to increase the accuracy of prediction and classification by maximizing the advantages of the ensemble theory by adding randomness to variables to obtain stability.

The random forest model performs N sampling data sets (i.e., N predictions) by randomly sampling a subset N times from a given data set through bagging (bootstrap aggregation) through a bootstrapping process on the data. Model). Subsequently, the process of selecting a random variable from each sampled data set proceeds.When selecting the number of variables, randomly selects as many as sqrt(M) or M/3 number of variables among the M total variables. Repeat the process of removing all remaining variables. In this way, an ensemble model is created by synthesizing decision trees in which variable selection has been performed, and misclassification is evaluated through an out-of-bag error.

When the random forest model is applied, the bias constituting the training error (if this value is high, the predicted result is inaccurate compared to the actual result), while the variance (if this value is high, the predicted result is It fits well in sets, but not in other data sets).

In one exemplary embodiment, when the central nucleotide of the window is a nucleotide that binds to a protein, the feature vector encoding the window may be defined as positive. On the other hand, when the window does not contain a nucleotide that binds to a protein, the feature vector encoding the window may be defined as negative. Feature vectors that are neither positive nor negative are excluded and may not be used in training and learning processes.

On the other hand, the sequence discovery means 230 filters a random sequence, and applies a random forest model trained through the above-described sequence learning means 220, and is a candidate RNA that binds to a target protein molecule among the filtered RNA sequences. Build and generate aptamer.

When the sequence discovery means 230 filters a random RNA sequence, the secondary structure, free energy, the number of unpaired bases, the same secondary structure in the pool as constraints of the secondary structure of the RNA sequence You can set the number of

As the secondary structure of the RNA sequence, the terminal portion (for example, the 1st to 5th bases of the RNA sequence and the 1st to 5th bases from the end) is the dangling end (two nucleic acid strands bonded to form a double strand) One end of the nucleic acid strand may or may not have a base longer or shorter than the end of the other nucleic acid strand, but the end portion may or may not have at least 3 base pairs (e.g. 3 to 7, preferably 3 to 5). Base pair).

The free energy of the RNA sequence should be -4.0 to -7.0 kcal/mol, preferably less than -5.0 to -6.0 kcal/mol. This takes into account the Z-score value of the free energy of known RNA aptamers.

Meanwhile, as a constraint for filtering, it is necessary to set the number of unpaired nucleotides in the RNA sequence to a certain number or more. In general, in the secondary structure of RNA, nucleotides in the single-stranded portion (i.e., nucleotides in the non-base pairing portion) frequently bind to proteins compared to nucleotides in the double-stranded portion (i.e., nucleotides in the base pairing portion) in the secondary structure of RNA. . There are several reasons why the single-stranded portion of nucleotides primarily bind to proteins. Because the nucleotides of the single-stranded portion can easily change their conformation, the nucleotides of the single-stranded portion (i.e., nucleotides that do not form a pair) are attached to the nucleotides of the double-stranded portion (i.e., the nucleotides forming the base pair). It is more flexible than that. In addition, nucleotides that do not form a pair have an electron donor or acceptor atom capable of forming a hydrogen bond with an amino acid of a protein.

Therefore, it may be desirable to set the number of nucleotides that do not form a pair that can easily form hydrogen bonds with amino acids of a protein while having a relatively flexible structure to be approximately 30% or more of the candidate RNA aptamer to be constructed. . For example, when designing 20 to 40 nucleotides of the candidate RNA aptamer, the number of unpaired nucleotides may be set to approximately 5 to 20 nucleotides.

In addition, when the same secondary structure is repeated in a pool, it may be difficult to construct an appropriate candidate RNA aptamer. Thus, the number of identical secondary structures in the pool can be set to less than 300, for example less than 200, preferably less than 150, for example less than 100.

The sequence discovery means 230 may predict, construct, and generate candidate RNA by applying the random forest model trained in the sequence learning means 230 to the filtered RNA sequence.

Thus, from known RNA-protein complexes, at their sequence level and structural level, the key features of the interacting RNA sequence and the target protein molecule are identified. In the present invention, a random forest (RF) model is constructed using the main characteristics of the interacting RNA and the target protein molecule. As confirmed later, the results of cross validation and independent testing of the RF model constructed in the present invention show that the series of potential RNA aptamers found in the RF model of the present invention are strong candidate aptamer sequences. Including, it is possible to greatly reduce the time and cost required by a process such as a typical SELEX.

Optionally, the structure of the candidate RNA aptamer predicted, constructed, and selected through the sequence learning means 220 and the sequence discovery means 230 according to the procedure schematically shown in FIG. 2 is analyzed using known structure analysis software. Can be analyzed. For example, using structure analysis software, the secondary structure of the candidate RNA aptamer selected by the sequence discovery means 230 and the binding structure centering on the binding site of the candidate RNA aptamer docked to the target protein molecule Can be analyzed.

At this time, using the structure analysis software, the secondary structure of the candidate RNA aptamer selected by the sequence discovery means 230 and the secondary structure of the known RNA aptamer can be compared and analyzed and/or candidates for the target protein molecule The binding structure of the candidate RNA aptamer to the target protein molecule and the known structure of the binding site of the candidate RNA aptamer to the target protein molecule in the state in which the RNA aptamer is docked to form a three-dimensional protein-RNA complex. The binding structure of the RNA aptamer to the target protein molecule can be compared.

For example, for the secondary structure of a candidate RNA aptamer and/or a known RNA aptamer, software such as RNAfold (Lorenz et al, 2011, Vienna RNA Package 2.0. Algorithms Mol. Biol. 6(1), 26) can be used. And, it can be visualized using software such as PseudoViewer (Byun and Han, 2009, PseudoViewer3: Bioinformatics 25(11), 1435-1437). Meanwhile, the binding structure of the candidate RNA aptamer and/or known RNA aptamer in the state of being docked on the target protein molecule is HDOCK (Yan et al., 2017, HDOCK: a web server for protein protein and proteinDNA/RNA docking based on a Hybrid strategy, Nucleic Acids Res. 45(W1), W365-W373) can be used, and the tertiary structure of RNA aptamer is RNAComposer (Popenda et al., 2012, Automated 3D structure composition for large RNAs, Nucleic Acids Res. 40(14), e112) can be used for analysis.

On the other hand, the candidate RNA aptamer generation program 210 may further include a sequence evaluation means 240 for verifying the random forest model used to generate the candidate RNA sequence by the sequence discovery means 230, if necessary. The sequence evaluation means 240 evaluates and verifies the positive (+) instances and negative (-) instances used in the random forest model of the sequence learning means 220 and the sequence discovery means 230 described above using an appropriate evaluation scale. do. If necessary, independent tests can also be performed by means of sequence evaluation 240.

In an exemplary embodiment, the sequence verification means 250 may be subjected to a standard cross-validation, for example a standard 10-fold cross-validation. Optionally, in addition to the standard 10-fold cross-validation, a leave-one-out (LOO) cross-validation may be performed. The reason for performing the LOO cross-validation is that the conventional k-fold cross-validation tends to overestimate the predictive performance for paired input values such as PPI (protein-protein interaction) or RNA interaction ( Abbasi, WA, Minhas, FUAA: Issues in performance evaluation for host-pathogen protein interaction prediction, Journal of Bioinformatics and Computational Biology 14(3):1650011 (2016)).

In one exemplary embodiment, the sequence evaluation means 240 is used as an evaluation measure for the random forest model used in accordance with the present invention as sensitivity, specificity, accuracy, and positive predictive value. predictive value, PPV), negative predictive value (NPV), and Matthews correlation coefficient (MCC), etc. may be used. These evaluation scales can be expressed by the following formulas (1) to (6).

In the above equations (1) to (6), the sensitivity is the ratio of the instances predicted to be positive by the random forest model among the actual positive instances of the sequence discovery means 230, and the specificity is the ratio of the actual negative instances by the random forest model. It is the ratio of the instances predicted to be negative, the accuracy is the ratio of the positive and negative instances predicted by the random forest model among the actual instances, and the positive predictive degree is the actual positive instance among the instances predicted to be positive by the random forest model. Is the ratio of, and the speech prediction is a measure of the ratio of the actual negative instances among the negatively predicted instances by the random forest model.

In addition, in Equations (1) to (6), TP (true positive) is a positive instance correctly predicted as a positive, TN (true negative, true negative) is a negative instance correctly predicted as a negative, FP (false positive, False positive) means a negative instance incorrectly predicted as positive, and FN (false negative, false negative) means a positive instance incorrectly predicted as negative.

If necessary, the sequence evaluation means 240 may perform an independent test for the candidate RNA construction model (RF model) identified in the present invention by using the data set identified in the previous study.

Meanwhile, the present invention relates to a method of generating a candidate RNA aptamer that binds to a target protein molecule by using the above-described RNA aptamer generation program 210 (see FIG. 1). 3 is a flow chart schematically showing a method of generating a candidate RNA aptamer that binds to a target protein molecule according to an exemplary embodiment of the present invention.

As shown in Figure 3, the method of generating a candidate RNA aptamer binding to a target protein molecule using a program realized in a computer is based on the protein-RNA complex data set, the RNA sequence and the amino acid sequence constituting the protein. Predict candidate RNA aptamers that bind to target protein molecules by using the random forest model to train and learn (step S210), filter the random RNA sequence, and use the trained random forest model from the filtered random RNA sequence. , Construction, including the step of generating (step S220), and if necessary, may include a step (step S230) of predicting, selecting, and evaluating the constructed candidate RNA aptamer using a scale.

Training and learning the RNA sequence and the amino acid sequence of the protein (step S210) may be performed by a sequence learning means (220, see FIG. 1). Characterization of the RNA sequence based on the protein-RNA complex data set identified through an experimental analysis method, for example, an analysis method such as CLIP and/or SELEX, and the amino acid sequence constituting the protein A vector can be constructed and the RNA sequence and protein sequence can be trained by applying a random forest model to the constructed feature vector. As an example, a feature vector for training an RNA sequence may include a tendency to interact with amino acids constituting a protein, a single-base composition of the RNA sequence, a 2-base composition and a pseudo 3-base composition of the RNA sequence. For example, a pseudo three-base composition can include the hydrophobicity, hydrophilicity of the nucleotides that make up the RNA sequence and the side-chain-weight of the base. At this time, the number of nucleotides of the trained RNA sequence may be set to 20 to 40, but the present invention is not limited thereto.

In addition, the feature vector for training the protein sequence may include composition-transition-distribution characteristics of amino acids constituting the protein and similar amino acid composition. The similar amino acid composition may include the hydrophobicity, hydrophilicity, side chain-weight, ionization index and isoelectric point of the amino acids constituting the protein.

Predicting, constructing, and generating a candidate RNA aptamer (step S220) may be performed by sequence discovery means 230 (see FIG. 1). In this step, the random forest model trained in the step of filtering the RNA sequence that meets a specific condition among the random RNA sequences and training and learning the sequence (step S210) is applied to the filtered RNA sequence to target the filtered RNA sequence. Candidate RNA aptamers that bind to protein molecules can be predicted, constructed, generated, and selected.

In an exemplary embodiment, when filtering a random RNA sequence, the random RNA sequence is limited by limiting the secondary structure of the random RNA sequence, the free energy, the number of unpaired nucleotides, and the number of identical secondary structures in the pool. RNA sequences can be filtered.

If necessary, the secondary structure and/or the binding structure in the state of docking with the target protein molecule can be analyzed using known structure analysis software for the selected and constructed candidate RNA aptamer. At this time, the selected candidate RNA aptamer, the secondary structure of the known RNA aptamer and the binding structure to the target protein molecule can be compared and analyzed.

Meanwhile, the step of evaluating the candidate RNA aptamer using an appropriate evaluation scale (step S230) may be performed by the sequence evaluation means 240 (see FIG. 1). In step S230, the random forest learning model used to predict, generate, and select candidate RNA is evaluated and verified using an appropriate evaluation scale. In an exemplary embodiment, the evaluation measures are Sensitivity, Specificity, Accuracy, Positive predictive value, Negative predictive value, and Matthews correlation coefficient. ) May be at least one selected from. The evaluation method is not particularly limited, but a standard cross-validation, for example, a standard 10-fold cross-validation or a Leave One (LOO) cross-validation may be performed. If necessary, an independent test using the results of previous studies can also be performed at this stage.

Hereinafter, the present invention will be described through exemplary embodiments, but the present invention is not limited to the technical idea described in the following examples.

Example

[data set]

Protein-RNA complexes elucidated by X-ray crystallography having a good resolution of 5.0 Å or more were extracted from the Protein Data Bank (PDB). Complexes with less than 10 short RNA sequences and more than 120 long RNA sequences were removed, leaving only 696 protein-RNA complexes in total. In order to learn the interaction propensity (IP) of nucleotides with amino acids, the remaining 696 protein-RNA complexes were used.

In addition, structural data for known RNA aptamer-protein complexes that are not included in the aforementioned 696 protein-RNA complexes were obtained from PDB. There were 35 RNA aptamer-protein complexes in the PDB.

In order to compare the model used in this example with other models, Li et al. from the aptamer base (Cruz et al., 2012, Aptamer Base: a collaborative knowledge base to describe aptamers and SELEX experiments.Database (Oxford), bas006). (Li et al., 2014, Predictor of Aptamer-Target Interacting Pairs with Pseudo-Amino Acid Composition, PLoS One 9(1), e86729), a benchmark dataset was obtained. As a positive example for testing the model according to this example, 56 RNA aptamer-protein pairs from positive data were used.

[Features for learning predictive models]

The SELEX procedure selects an aptamer from an oligo-nucleotide library having various regions of 30-mer or 60-mer, the aptamer selected is usually shorter than 30-mer. Therefore, in this example, a 27-mer RNA aptamer was designed to be generated in order to generate a potential RNA aptamer consisting of a nucleic acid sequence shorter than a 30-mer binding to a target protein molecule. For each pair of RNA and protein sequences, the following characteristics were encoded.

RNA characteristics: IP of nucleotide triplets with amino acids, mono-nucleotide composition (mC, single-base composition), di-nucleotide composition (dC, 2-base composition), similar tri-nucleotide composition (PseTNC, pseudo 3 -Base composition)

Protein characteristics: composition-transition-distribution of amino acid groups (C-T-D), similar amino acid composition (PseAAC).

To define the IP of the nucleotide triplet with the amino acid, a data set of 696 protein-RNA complexes was used to calculate the IP of 1,280 (64 X 20) between the nucleotide triplet of 4 ³ =64 and 20 amino acids. . In addition, for each RNA sequence, the single-base composition (mC) and the 2-base composition (dC) were also calculated.

Meanwhile, in order to express the protein sequence, 20 amino acids were clustered into 7 groups based on their dipoles and volume. {M (methionine), S (Serine), T (Threonine), Y (Tyrosine)}, {F (phenylalanine), I (Isoleucine), L (Leucine), P (Proline)}, {H (Histidine), N (Asparagine), Q (Glutamine), W (Tryptophan)}, {A (Alanine), G (Glycine), V (Valine)}, {K (Lysine), R (Arginine)}, {D (Aspartic acid ), E (Glutamic acid)}, {C (Cysteine)}. For each group of amino acids in the protein sequence, a composition-transition-distribution (CTD) feature was expressed, which required 7 + 21 + 35 = 63 elements in the feature vector (Dubchak et al. ., 1995, Prediction of protein folding class using global description of amino acid sequence, Proc. Natl. Acad. Sci. USA 92(19), 8700-8704).

As an additional feature, a pseudo trinucleotide composition (PseTNC) for each RNA sequence (Chen et al., 2014, it is-PseTNC: a sequence-based predictor for identifying initiation site in human genes using pseudo trinucleotide composition.Anal. Biochem. 462, 76-83) and similar amino acid composition for each protein sequence (PseAAC) (Shen and Chou, 2008. PseAAC: a flexible web server for generating various kinds of protein pseudo amino acid composition, Anal. Biochem. 373 (2), 386-388) were calculated, these features taking into account the sequence-order effect of residues within the sequence.

Specifically, with regard to PseTNC, PseTNC was calculated using three physiochemical properties of nucleotides, namely, hydrophobicity, hydrophilicity, and side-chain mass. Similarly, with respect to PseAAC, various physiochemical properties of amino acids, namely hydrophobicity, hydrophilicity, side chain-weight, ionization index 1 (pK1, a-CO ₂ H), ionization index 2 (pK2, NH ₃ ), at 25°C. PseAAC was calculated using the isoelectric point (pi) of. When calculating PseTNC and PseAAC, the weight factor (w) was set to 0.1 and the correlation tier, λ, was set to 1. PseTNC and PseAAC represent tri-nucleotide composition and amino acid composition each having a tier correlation factor. Since the number of tier correlation factors is equal to the parameter λ, the number of elements in the feature vectors for PseTNC and PseAAC are 65 and 21, respectively. In a feature vector with 669 elements, a pair of RNA windows and protein pairs are encoded: 500 IP, 4 mC, 16 dC, 65 PseTNC, 63 CTD, 21 PseAAC.

[Positive and negative instances for training]

A random forest (RF) model was constructed, and the RF model was trained on the aforementioned feature vector generated from the protein-RNA complex. When constructing the RF model, the number of trees was set to 35, and the number of features was set as the square root of the number of feature elements.

Although generating potential RNA aptamers of less than 30-mers is the primary aim of the present invention, RNA sequences have varying lengths when training and testing data sets. As shown in Fig. 3, a positive instance and a negative instance were defined using 27 nucleotides. As the RNA sequence is scanned using a window of 27 nucleotides, a feature vector representing the window is created. If the central nucleotide of the window was a protein-binding nucleotide, then the feature vector encoding the window was considered positive. If a window contains only protein-unbound nucleotides, the feature vector for that window is considered negative. Feature vectors that are neither positive nor negative have been removed from the training data set because they result in very unbalanced positive/negative instances for training. However, as an exception, both the first and last windows were considered positive if they contained protein-binding nucleotides at any position in the window. In the training dataset according to this example, the ratio of positive instances to negative instances was about 3:1.

[Structural restriction of RNA aptamer]

In order to discover a 27-mer RNA aptamer for a given protein sequence, 6 X 10 ⁶ 27-mer random RNA sequences were generated, and RNAfold (Lorenz et al., 2011. Vienna RNA Package 2.0. Algorithms Mol. Biol. 6) (1) and 26) were used to predict their secondary structure. Among the random RNA sequences, those in which the secondary structure satisfies the following four constraints were filtered, and an initial pool of candidate aptamers was constructed.

First, the secondary structure must be terminated by at least three consecutive bases with or without dangling end (terminal longer or shorter than the dangling end) base in one of the following forms. The form is (((N21))),.(((N20))),(((N20))). and .(((N19))). Where Nk is a k-mer sequence fragment.

Second, the free energy should be less than -5.7 kcal/mol.

Third, the secondary structure should contain at least 11 unpaired bases.

Fourth, the number of identical secondary structures in the pool should not exceed 150.

The above four constraints are derived from the analysis of all known RNA aptamer-protein complexes. As shown in Fig. 4, the secondary structure of most RNA aptamers having a known structure terminates with a stem having at least three base pairs with or without dangling terminal bases.

The free energy constraint was obtained from a study conducted by Chushak et al. (Chushak and Stone, 2009, In silico selection of RNA aptamers. Nucleic Acids Res. 37(12), e87). The researchers calculated the Z-score and standard deviation using the mean free energy of 10,000 equal-length random sequences. Since the Z-score of the free energy of known RNA aptamers is in the range of -0.9 to -4.3, the cut-off value of the free energy is -5.7 kcal/corresponding to the Z-score -1. set to mol.

For example, as shown in Figure 4, in the structure of a known RNA aptamer that binds to a target protein molecule, nucleotides that bind to the protein are frequently observed in a single-stranded portion rather than a double-stranded portion of the RNA secondary structure. (Chushak and Stone, 2009; Carothers et al., 2004, Informational complexity and functional activity of RNA structures. J. Am. Chem. Soc. 126(16), 5130-5137).

In consideration of the length of the RNA aptamer, candidate aptamers having at least 11 unpaired nucleotides in this Example were selected (3rd constraint).

Many secondary structures formed by the initially generated 6 X 10 ⁶ random RNA sequences are mutually similar. Some secondary structures were observed over 10,000 times, while others were observed only once. To maintain the diversity of potential RNA aptamers, the number of identical secondary structures was limited to 150 (4th constraint).

After applying the above four structural constraints, only 38,327 RNA sequences remained from the first 6 X 10 ⁶ random sequences. For each protein target molecule, a pair of RNA-protein sequences was encoded in a feature vector for the RF model described above. The RF model consists of 35 sub-trees, each sub-tree being voted on an input feature vector. For all RNA sequences screened by the above four constraints, the probability is estimated from the ratio of positive votes by calculating with the RF model applied when training the RNA sequence and the amino acid sequence of the protein. estimate). In relation to free energy and probability, 38,327 RNA sequences selected under constraints were ranked, and among the sequences with the highest probability, candidate RNA aptamers having the lowest free energy ranking were selected.

[Cross Verification and Independent Test]

A learning model for predicting and generating the selected candidate RNA was evaluated by performing 10-fold cross validation, leave-one-out cross validation, and independent testing. Cross-validation and independent testing are the sensitivity (Sn), specificity (Sp), accuracy (Accuracy, Acc), positive predictive value, PPV defined by the above equations (1) to (6). ), negative predictive value (NPV), and Matthew's correlation coefficient (MCC). Table 1 shows the results of 10-fold cross-validation and leaving one cross-validation related to constructing and generating a candidate RNA aptamer that binds to a target protein molecule according to this Example. As shown in Table 1, both of the two cross-validations showed excellent performance, and in particular, compared to the one left cross-validation, the 10-fold cross-validation showed better performance.

Validation of candidate RNA aptamer construction model Verification Sn Sp Acc PPV NPV MCC 10-fold crossing 95.10% 98.69% 97.79% 96.04% 98.37% 0.941 LOO 94.12% 98.04% 97.06^ 94.12% 98.04% 0.922

On the other hand, in the independent test for the RF model constructed according to this example, other previous studies (Li et al., 2014, Prediction of Aptamer-Target Interacting Pairs with Pseudo-Amino Acid Composition.PLoS One 9(1), e86729 ; An RNA aptamer binding to a protein from the benchmark data set of Zhang et al., Prediction of aptamer-protein interacting pairs using ensemble classifier in combination with various protein sequence attributes.BMC Bioinformatics 17, 225) was used as a positive instance. Previous studies used the same benchmark data set. In previous studies, negative instances were generated by randomly pairing aptamers and protein targets in the positive data set, but in previous studies, when the protein partner changes to one of the protein targets in the positive data set, the negative instance becomes a positive instance. I did not use negative instances of the study. For negative instances associated with independent testing, random RNA sequences were generated and paired with one of the protein chains in 35 protein-RNA complexes obtained from the PDB. The results of the independent test are shown in Table 2 below.

Independent testing of candidate RNA aptamer construction models Way Sn Sp Acc PPV NPV MCC Li et al. 48.3% 87.1% 77.4% - - 0.372 Zhang et al. 73.8% 71.3% 71.9% - - 0.388 Example 76.8% 66.1% 71.4% 69.4% 74.0% 0.431

As shown in Table 2, the RF model of the Example for constructing a candidate RNA aptamer binding to a target protein molecule according to this Example generally showed superior performance compared to previous papers. Due to the difference between the preceding paper and the negative instance used in this example, sensitivity seems to be the most important and fair evaluation measure for comparison.

[conclusion]

Conventional methods of combining RNA aptamers with target protein molecules using computer operations cannot be used to discover new RNA aptamers for target protein molecules. This is because the model adopted in the prior art is limited to specific targets, and is a classification model that determines whether a given pair of proteins and RNA sequences interact. In this example, we propose a computer model for constructing a potential candidate RNA aptamer for a target protein molecule using various features of interacting RNA and protein, and structural constraints on the secondary structure of RNA. A random forest model was constructed using various features of RNA and protein sequences, and the sequences were trained and learned. A random RNA sequence can be filtered according to an appropriate criterion, and a candidate RNA aptamer for a target protein molecule can be predicted and constructed using the constructed random forest model. The constructed candidate RNA aptamer showed excellent performance in cross-validation and independent tests. Therefore, since the size of the pool of initial nucleic acid sequences binding to the target protein molecule can be greatly reduced by applying the model of this example, it is expected that enormous time and cost required for in vitro experiments can be saved. .

In the above, the present invention has been described based on exemplary embodiments and examples of the present invention, but the present invention is not limited to the technical idea described in the above embodiments and examples. Rather, a person of ordinary skill in the art to which the present invention pertains can easily add various modifications and changes based on the above-described embodiments and examples. However, it is clear from the appended claims that these modifications and changes all belong to the scope of the present invention.

100: computer 110: input
120: output unit 130: display
140: memory 150: drive/application
160: control unit
200: recording medium
210: candidate RNA aptamer generation program
220: sequence learning means 230: sequence discovery means
240: sequence evaluation means

Claims (10)

Based on the RNA-protein complex data, the feature vector of the RNA sequence and the feature vector of the amino acid sequence constituting the protein are constructed, and the RNA sequence and the protein sequence are trained by applying a random forest model based on the constructed feature vector. Sequence learning means to let; And
Filtering the random RNA sequence, and applying the random forest model learned through the sequence learning means, comprising a sequence discovery means for constructing a candidate RNA aptamer binding to the target protein molecule among the filtered RNA sequences,
In the sequence learning means, the RNA feature vector has an interaction propensity with amino acids constituting a protein, a mono-nucleotide (single-base) composition of the RNA sequence, and a di-nucleotide 2 -Base) composition and pseudo tri-nucleotide (pseudo tri-nucleotide) composition,
In the sequence learning means, the protein feature vector is a computer-readable record recording a program for generating a candidate RNA aptamer that binds to a target protein molecule created based on the composition, transfer and distribution of amino acids constituting the protein, and similar amino acid composition. media.
The method of claim 1,
The program includes a program for generating a candidate RNA aptamer that binds to a target protein molecule further comprising a sequence evaluation means for evaluating a model for generating a candidate RNA aptamer constructed by the sequence discovery means using an evaluation scale. A recording medium that can be read by a recorded computer.
The method of claim 2,
The evaluation scale is at least selected from sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and Matthews correlation coefficient. A computer-readable recording medium in which a program for generating a candidate RNA aptamer binding to a target protein molecule, characterized in that one is one, is recorded.
The method according to any one of claims 1 to 3,
The sequence discovery means filters the random RNA sequence by limiting the secondary structure of the random RNA sequence, free energy, the number of unpaired nucleotides, and the number of identical secondary structures in a pool. A computer-readable recording medium in which a program for generating a candidate RNA aptamer that binds to the target protein molecule described above is recorded.
The method according to any one of claims 1 to 3,
The similar tri-nucleotide composition constituting the RNA characteristic vector includes hydrophobicity, hydrophilicity and side-chain mass of the tri-nucleotide,
The similar amino acid composition constituting the protein characteristic vector can be read by a computer recording a program for generating a candidate RNA aptamer that binds to a target protein molecule including the hydrophobicity, hydrophilicity, side chain-weight, ionization index, and isoelectric point of the amino acid. Capable of recording media.
As a method of generating a candidate RNA aptamer that binds to a target protein molecule using a computer-implemented program,
By means of sequence learning, a feature vector of an RNA sequence and a feature vector of an amino acid sequence constituting a protein are constructed based on the RNA-protein complex data, and a random forest model is applied based on the constructed feature vector. Training the sequence and protein sequence; And
Filtering the random RNA sequence by means of sequence discovery, and applying the random forest model learned through the sequence learning means to construct a candidate RNA aptamer that binds to the target protein molecule among the filtered RNA sequences Including,
In the sequence learning means, the RNA feature vector has an interaction propensity with amino acids constituting a protein, a mono-nucleotide (single-base) composition of the RNA sequence, and a di-nucleotide 2 -Base) composition and pseudo tri-nucleotide (pseudo tri-nucleotide) composition,
In the sequence learning means, the protein feature vector is a method of generating a candidate RNA aptamer that binds to a target protein molecule created based on the composition, transfer and distribution of amino acids constituting the protein, and similar amino acid composition.
The method of claim 6,
A method of generating a candidate RNA aptamer binding to a target protein molecule, further comprising the step of evaluating a model for generating a candidate RNA aptamer constructed by the sequence discovery means by a sequence evaluation means using an evaluation scale.
The method of claim 7,
The evaluation scale is at least selected from sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and Matthews correlation coefficient. Method for producing a candidate RNA aptamer binding to a target protein molecule, characterized in that one.
The method according to any one of claims 6 to 8,
The sequence discovery means filters the random RNA sequence by limiting the secondary structure of the random RNA sequence, free energy, the number of unpaired nucleotides, and the number of identical secondary structures in a pool. A method of generating a candidate RNA aptamer that binds to the target protein molecule.
The method according to any one of claims 6 to 8,
The similar tri-nucleotide composition constituting the RNA characteristic vector includes hydrophobicity, hydrophilicity and side-chain mass of the tri-nucleotide,
The similar amino acid composition constituting the protein characteristic vector is a method of generating a candidate RNA aptamer that binds to a target protein molecule including hydrophobicity, hydrophilicity, side chain-weight, ionization index, and isoelectric point of the amino acid.

Images