Computer system and method for training prediction model

Abstract

This computer system: connects to a database storing read results for the base sequences of a plurality of biopolymers and disintegration rates corresponding to the biopolymers; generates, for each of the biopolymers, data representing depth, which is the number of overlaps of read results at each position in the base sequence of the biopolymer; calculates, for each of the biopolymers, the standard deviation of the depths on the basis of the data; classifies the biopolymers into groups on the basis of the standard deviation of the depths; generates, for each of the groups, training data configured from the base sequences and the disintegration rates of the biopolymers; and executes, for each of the groups, machine learning in which the training data is used, thereby generating a prediction model that receives input of the base sequence of a biopolymer and outputs the disintegration rate of the biopolymer.

Description

Computer system and predictive model learning method Incorporation by Reference

[0001] This application claims priority from Japanese Patent Application No. 2024-066820, filed on April 17, 2024, the contents of which are incorporated herein by reference.

[0002] The present invention relates to a system and method for predicting the stability of biopolymers.

[0003] Biopolymers are polymers present in living organisms and are composed of polymerized monomer units. Known examples of biopolymers include polynucleotides, polypeptides, and polysaccharides. Because biopolymers are known to have diverse correlations between their polymerization structure and physiological activity, computers are often used to evaluate the structure and efficacy of biopolymers when developing them as pharmaceuticals. For example, it is known that the stability of biopolymers against degradation in the body can be predicted.

[0004] Non-Patent Document 1 describes how the stability of mRNA as a biopolymer is defined as half-life, how the half-lives of mRNAs of many genes are measured by RNA-Seq, and how the half-lives are then used for machine learning to predict the half-life of mRNA of any sequence.

[0005] Medina-Munoz, Santiago Gerardo, et al, “Crosstalk between codon optimality and cis-regulatory elements dictates mRNA stability,” Genome biology 22.1 (2021):1-23

[0006] The method described in Non-Patent Document 1 has a problem in that the correlation coefficient between the predicted value of mRNA stability and the actual measured value is small. An object of the present invention is to provide a system that can predict the stability of biopolymers with high accuracy.

[0007] A representative example of the invention disclosed in the present application is as follows: That is, a computer system includes a processor, a storage device connected to the processor, and a network interface connected to the processor, and is accessible to a database that stores read results of base sequences of a plurality of biopolymers and decay rates corresponding to the biopolymers, wherein the processor generates, for each of the biopolymers, data representing a depth, which is the number of overlaps of the read results at each position in the base sequence of the biopolymer, calculates, for each of the biopolymers, a standard deviation of the depth of the biopolymer based on the data, classifies the biopolymers into a plurality of groups based on the standard deviation of the depth of each of the biopolymers, generates, for each of the groups, training data consisting of the base sequences of the biopolymers belonging to the group and the decay rates stored in the database, and performs machine learning using the training data of the group for each of the groups to generate a prediction model that accepts the base sequences of the biopolymers and outputs the decay rates of the biopolymers, and stores information about the prediction model in the storage device.

[0008] According to the present invention, the stability of any biopolymer can be predicted with high accuracy. Objects, configurations, and effects other than those described above will become apparent from the following description of the examples.

[0009] FIG. 1 is a block diagram showing an example of the configuration of a computer system according to Example 1. FIG. 2 is a block diagram showing an example of the hardware configuration of a server according to Example 1. FIG. 3 is a diagram showing an example of a GUI of the computer system according to Example 1. FIG. 4 is a diagram showing an example of a GUI of the computer system according to Example 1. FIG. 5 is a flowchart explaining an example of a prediction model generation process executed by the server according to Example 1. FIG. 6 is a diagram explaining an example of depth data generated by a depth calculation unit according to Example 1. FIG. 7 is a diagram explaining an example of a database created by the server according to Example 1. FIG. 8 is a flowchart explaining an example of a prediction process executed by the server according to Example 1. FIG. 9 is a diagram showing an example of a GUI of the computer system according to Example 1. FIG. 10 is a diagram showing an example of a verification result of the performance of a prediction model according to the present invention. FIG. 11 is a diagram showing an example of a verification result of the performance of a prediction model according to the present invention.

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the present invention should not be construed as being limited to the description of the embodiments shown below. Those skilled in the art will readily understand that the specific configuration can be changed without departing from the concept or spirit of the present invention.

[0011] In the configuration of the invention described below, the same or similar configurations or functions are denoted by the same reference numerals, and redundant explanations will be omitted.

[0012] In this specification, the terms "first," "second," "third," etc. are used to identify components and do not necessarily limit the number or order.

[0013] To facilitate understanding of the invention, the position, size, shape, range, etc. of each component shown in the drawings etc. may not represent the actual position, size, shape, range, etc. Therefore, the present invention is not limited to the position, size, shape, range, etc. disclosed in the drawings etc.

[0014] Fig. 1 is a block diagram showing an example of the configuration of a computer system according to Example 1. Fig. 2 is a block diagram showing an example of the hardware configuration of a server according to Example 1.

[0015] The computer system is made up of a server 100 and a client terminal 101. The server 100 and the client terminal 101 are connected via a network such as a LAN (Local Area Network).

[0016] The server 100 trains a prediction model that predicts an index for evaluating the stability of a biopolymer and predicts the index using the prediction model. As shown in FIG. 2 , the server 100 has a processor 200, a network interface 201, a main memory device 202, and a secondary memory device 203. The hardware elements are connected to each other via a bus 204. The server 100 may also have input devices such as a keyboard, a mouse, a touch panel, etc., and an output device such as a display.

[0017] Information held by the server 100 is stored in the main memory device 202 or the secondary memory device 203. Furthermore, the functional units of the server 100 are realized by the processor 200 executing programs stored in the main memory device 202 or the secondary memory device 203. In the following, when a process is described using a functional unit as the subject, it indicates that the processor 200 is executing a program that realizes the functional unit.

[0018] The functional units realized by the programs may be realized using dedicated circuits such as a virtual computer, a computer system, an FPGA (Field-Programmable Gate Array), and an ASIC (Application Specific Integrated Circuit).

[0019] In the following, an example of a biopolymer will be described, taking mRNA, which is a type of RNA that is a polynucleotide, as an example.

[0020] Non-Patent Document 1 defines the mRNA decay rate as an index for evaluating mRNA stability. In the present invention, the mRNA decay rate is also used as an index for evaluating mRNA stability. However, although the decay rate is used as an example, other stability indicators may also be used.

[0021] The server 100 receives an input of an mRNA base sequence using mRNA base sequences and mRNA RNA-Seq data, and performs machine learning to generate a prediction model that predicts the decay rate of the mRNA. The prediction model in this embodiment predicts the decay rate of the mRNA base sequence. The prediction model may be, for example, a deep learning model such as a DNN, but is not limited to deep learning.

[0022] The server 100 holds an RNA-Seq data database 110 and a prediction model database 130.

[0023] The RNA-Seq data database 110 is a database that stores RNA-Seq data for various mRNAs. RNA-Seq data is the result of reading base sequences obtained using next-generation sequencing. For example, it includes a read ID, read sequence, and quality. The RNA-Seq data is managed in association with the elapsed time (time series) from the measurement reference time. The RNA-Seq data database 110 may also store data in association with the decay rates corresponding to various mRNAs. The RNA-Seq data database 110 may be maintained by a device or system external to the server 100.

[0024] The prediction model database 130 is a database that stores information that defines a prediction model.

[0025] The server 100 has a depth calculation unit 120, a standard deviation calculation unit 121, an mRNA classification unit 122, a decay rate calculation unit 123, a learning unit 124, and a prediction unit 125. The functional units of the server 100 may be configured such that multiple functional units are combined into one functional unit, or one functional unit may be divided into multiple functional units for each function.

[0026] The client terminal 101 inputs various data to the server 100 and acquires various data from the server 100. The hardware configuration of the client terminal 101 is similar to that of the server 100, so a description thereof will be omitted. The client terminal 101 has an RNA-Seq data registration unit 140, an input data input unit 141, and a prediction result acquisition unit 142.

[0027] The RNA-Seq data registration unit 140 registers RNA-Seq data in the server 100. For example, as shown in FIG. 3, RNA-Seq data of biopolymers may be registered from the client terminal 101. In addition to the RNA-Seq data, species information related to the RNA-Seq data and measurement time may also be input. The input data input unit 141 inputs input data to be input into the prediction model to the server 100. For example, only information on biopolymers whose stability is to be predicted may be selected and input into the prediction model. Specifically, as shown in FIG. 4, only specific sequences may be selected from the RNA-Seq data using check buttons and input into the prediction model. The prediction result acquisition unit 142 acquires prediction results from the server 100. The client terminal 101 may also have a functional unit for inputting various machine learning parameters for the prediction model.

[0028] FIG. 5 is a flowchart illustrating an example of a prediction model generation process executed by the server 100 according to the first embodiment.

[0029] The depth calculation unit 120 generates depth data for each mRNA (step S101). Specifically, the following process is performed.

[0030] (S101-1) The depth calculation unit 120 selects one mRNA from the RNA-Seq data database 110.

[0031] (S101-2) The depth calculation unit 120 acquires RNA-Seq data for the selected mRNA from the RNA-Seq data database 110. Here, RNA-Seq data is acquired for each time series (each elapsed time). For example, RNA-Seq data for "0 hours," which is the initial measurement reference time for RNA-Seq of the selected mRNA, RNA-Seq data for "1 hour" elapsed since the initial measurement, RNA-Seq data for "3 hours" elapsed since the initial measurement, and RNA-Seq data for "6 hours" elapsed since the initial measurement are acquired.

[0032] (S101-3) The depth calculation unit 120 maps the RNA-Seq data for each time series to the reference sequence of the selected mRNA.

[0033] (S101-4) For each time series, the depth calculation unit 120 calculates the number of overlapping reads (mRNA amount) detected by the RNA-Seq data read at each position of the reference sequence based on the mapping results as the depth for each base, and saves it as depth data.

[0034] (S101-5) The depth calculation unit 120 determines whether the process of S101-4 has been completed for all of the mRNAs to be processed. If the process has not been completed for all of the mRNAs to be processed, the depth calculation unit 120 returns to S101-1. If the process has been completed for all of the mRNAs to be processed, the depth calculation unit 120 ends the process of S101.

[0035] 6 is a diagram illustrating an example of depth data generated by the depth calculation unit 120 according to the first embodiment. The horizontal axis represents the base sequence, and the vertical axis represents the depth (mRNA amount). FIG. 6 illustrates depth data for the same mRNA with different time series.

[0036] The greater the change in depth over time, the lower the stability of the mRNA, and the smaller the change in depth over time, the higher the stability of the mRNA.

[0037] The standard deviation calculation unit 121 calculates the standard deviation of depth for each mRNA using depth data of any time series (step S102). The standard deviation of depth is calculated for each mRNA having depth data. For example, the depth variation of the full-length mRNA sequence is calculated based on the depth data of the full-length mRNA sequence at the elapsed time "0 hours" of the mRNA shown in FIG. 6. Specifically, as shown in Table A of FIG. 7, the standard deviation of depth is calculated for each mRNA based on the depth data of the full-length mRNA sequence. A database of the calculated standard deviation of depth may be stored in the prediction model database 130. Hereinafter, in this embodiment, the depth data at the elapsed time "0 hours" is used. Furthermore, although the standard deviation is used in this embodiment, variance may also be used.

[0038] The mRNA classification unit 122 classifies the plurality of mRNAs into a plurality of groups based on the standard deviation of the mRNA amount of each mRNA (step S103). For example, the mRNA classification unit 122 classifies the plurality of mRNAs into eight groups based on the classification criteria and the standard deviation of the mRNA amount. The classification criteria are the range of the standard deviation of the mRNA amount and can be set arbitrarily. For example, classification criteria for classification into four groups may be set. Therefore, mRNAs having standard deviations falling within a predetermined range are classified into each group. Specifically, as shown in Table C of FIG. 7, a range of standard deviation of depth is set for each group. Furthermore, the plurality of mRNAs are classified into a plurality of groups by extracting mRNAs corresponding to the set range of standard deviation of depth. The mRNA classification results may be stored as a classification database in the prediction model database 130.

[0039] The decay rate calculation unit 123 calculates the decay rate of each mRNA (step S104). Specifically, the following process is executed. Note that if the decay rate corresponding to each mRNA is stored in advance, step S104 may be omitted.

[0040] (S104-1) The decay rate calculation unit 123 selects mRNA.

[0041] (S104-2) The decay rate calculation unit 123 divides each time series of depth data into a predetermined number of regions. A region is a predetermined range in the mRNA to be analyzed. The number of regions can be set arbitrarily. Note that a region is determined by the position i of the base sequence that indicates the start point of the region and the width w of the region.

[0042] (S104-3) The collapse speed calculation unit 123 selects an area.

[0043] (S104-4) The decay rate calculation unit 123 calculates the amount of change in depth of the selected region of each time series of depth data, and calculates the decay rate based on the amount of change. A possible method for calculating the decay rate is regression analysis based on a reaction model of mRNA decay. The decay rate calculation unit 123 stores the mRNA, region, and decay rate in association with each other.

[0044] (S104-5) The collapse speed calculation unit 123 determines whether or not the process has been completed for all regions. If the process has not been completed for all regions, the collapse speed calculation unit 123 returns to S104-3.

[0045] (S104-5) When processing has been completed for all regions, the decay rate calculation unit 123 determines whether processing has been completed for all mRNAs. If processing has not been completed for all mRNAs, the decay rate calculation unit 123 returns to S104-1. When processing has been completed for all mRNAs, the decay rate calculation unit 123 ends the processing of S104. As an example, when processing has been completed for all mRNAs, a database of decay rates may be generated as shown in Table B of FIG. 7. The decay rate database may be stored in the prediction model database 130 or the RNA-Seq data database 110. Although the method for calculating the decay rate for each region has been described above, the decay rate relative to the entire length of the target mRNA or the average decay rate for each mRNA region may also be calculated.

[0046] The learning unit 124 learns and generates a prediction model for each of the mRNA groups classified in step S103 (step S105). Specifically, the following process is executed.

[0047] (S105-1) The learning unit 124 selects a group.

[0048] (S105-2) The learning unit 124 generates learning data for each mRNA belonging to the group, consisting of the base sequence of the mRNA and the decay rate of the mRNA calculated in S104-4. Specifically, as shown in Table D in Figure 7, the learning unit 124 generates the base sequences and decay rates of the mRNA belonging to Group A based on the databases of Tables B and C in Figure 7.

[0049] (S105-3) The learning unit 124 uses the generated learning data to learn a prediction model that receives the base sequence of mRNA as input and outputs the decay rate of mRNA. The present invention is not limited to a learning method.

[0050] (S105-4) The learning unit 124 determines whether or not the processing has been completed for all groups. If the processing has not been completed for all groups, the learning unit 124 returns to S105-1.

[0051] (S105-4) When processing has been completed for all groups, the learning unit 124 associates the group identification information with the prediction model and stores them in the prediction model database 130. At this time, the learning unit 124 also stores the calculation results of the standard deviation of the mRNA amount and the classification criteria in the prediction model database 130.

[0052] In the present invention, RNA-Seq data was classified into multiple groups based on the standard deviation of mRNA abundance calculated from the mRNA RNA-Seq data, and a training dataset was generated for each group. Furthermore, in the present invention, a prediction model was trained to predict the decay rate of each biopolymer region, taking into account that the stability of biopolymers such as mRNA varies from region to region. This allows each prediction model generated using the training data from each group to demonstrate high prediction accuracy.

[0053] Figure 10 shows the results of generating depth data for multiple RNA-Seq data, classifying multiple mRNAs based on the standard deviation calculated from the depth data, and learning each group. The number of groups was set to two, four, or eight. The classification criteria were based on the number of genes contained in each group. Specifically, in the two-classification method shown in Figure 10, the number of genes contained in each group was approximately 2,000, in the four-classification method it was approximately 1,000, and in the eight-classification method it was approximately 500.

[0054] In addition, three types of learning methods were used: LASSO, CNN, and Enformer.

[0055] The correlation coefficient is the correlation between the predicted value of the decay rate, which is one of the stabilities of mRNA, and the actually measured value.

[0056] The results showed that the prediction accuracy of the four-category prediction model was high. In particular, the prediction model trained using the Enformer with four-category classification had an average accuracy of 0.9495.

[0057] Moreover, the prediction accuracy using other mRNA data sets is shown in FIG.

[0058] In Figure 11, the prediction model trained with Enformer, which was classified into eight categories, had high prediction accuracy. As can be seen, prediction accuracy varies depending on the number of genes included in the classified group. This is thought to be related to the fact that mRNA depth data depends on the secondary structure of the mRNA. If the number of categories is large, the distribution of SD for each group will be uniform, and the secondary structure of the mRNA included in the group will also be uniform, which is thought to improve learning performance. On the other hand, if the number of categories is larger than a certain range, the number of mRNA data for learning will decrease, and learning performance will likely decrease.

[0059] Furthermore, since the deep learning methods Enformer and CNN had higher prediction accuracy than Lasso, the use of deep learning makes it possible to accurately predict the decay rate of mRNA.

[0060] In this example, the prediction model in Example 1 is used to calculate the decay rate from the base sequence of mRNA or RNA-Seq data.

[0061] FIG. 8 is a flowchart illustrating an example of a prediction process executed by the server 100 according to the first embodiment.

[0062] The prediction unit 125 acquires input data from the client terminal 101 (step S201).

[0063] In this example, two possible patterns are input: one is to input only the mRNA base sequence, and the other is to input RNA-Seq data and mRNA base sequences with different time series. Therefore, the prediction unit 125 determines whether the input data includes RNA-Seq data (step S202).

[0064] If the input data includes RNA-Seq data, the prediction unit 125 calculates the standard deviation of the depth corresponding to the RNA-Seq data of the input data in cooperation with the depth calculation unit 120 and the standard deviation calculation unit 121 (step S203).

[0065] The prediction unit 125 refers to the classification criteria stored in the prediction model database 130 based on the standard deviation of the mRNA amount, and identifies the group to which the mRNA belongs (step S204).

[0066] The prediction unit 125 obtains information about the prediction model for the identified group from the prediction model database 130 and predicts the mRNA decay rate by inputting the mRNA base sequence into the prediction model (step S205). The prediction unit 125 then outputs the prediction result (mRNA decay rate) to the client terminal 101 (step S209) and terminates the prediction process. For example, FIG. 9 shows an example of a display screen on the client terminal 101. As shown in FIG. 9, the RNA-Seq data to be predicted may be received as input via the RNA-Seq data registration unit 140, and the standard deviation calculated by the prediction unit 125 may be output. Alternatively, the prediction model to be used may be received as input from the client based on the standard deviation. In FIG. 9, selection information for the prediction model to be used is received as input from the user via the client terminal 101. However, the prediction model to be used may be automatically selected based on the calculated standard deviation of the mRNA depth without user input. Finally, the decay rate of the mRNA to be predicted may be output to the client terminal 101 via the prediction result acquisition unit 142. The correlation function described in the prediction model to be used indicates the correlation between the predicted value of the collapse rate used in training each prediction model and the actual measured value of the collapse rate obtained through experiments or the like.

[0067] In step S202, if the input data does not include RNA-Seq data, the prediction unit 125 obtains information about the prediction model for each group from the prediction model database 130, and predicts the mRNA decay rate by inputting the mRNA base sequence of the input data into each prediction model (step S206).

[0068] The prediction unit 125 calculates the average value of the decay rate for each prediction model for each group (step S207). Although the average value is used as an example here, a statistical value such as a median may be used.

[0069] The prediction unit 125 selects a prediction result to output based on the average value of the decay rates (step S208). In this embodiment, the average value of the decay rates is selected. The prediction unit 125 then outputs the prediction result (mRNA decay rate) to the client terminal 101 (step S209), and the prediction process ends.

[0070] According to this embodiment, a prediction model suitable for prediction of the target mRNA can be selected, thereby enabling the decay rate of the target mRNA to be calculated with high accuracy.

[0071] The present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments are provided to explain the present invention in detail, and the present invention is not necessarily limited to those including all of the described configurations. Furthermore, some of the configurations of each embodiment can be added to, deleted from, or replaced with other configurations.

[0072] Furthermore, some or all of the above-described configurations, functions, processing units, processing means, etc. may be implemented in hardware, for example, by designing them as integrated circuits. The present invention can also be realized by software program code that implements the functions of the embodiments. In this case, a storage medium on which the program code is recorded is provided to a computer, and a processor included in the computer reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium implements the functions of the above-described embodiments, and the program code itself and the storage medium on which it is stored constitute the present invention. Examples of storage media for providing such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, solid-state drives (SSDs), optical disks, magneto-optical disks, CD-Rs, magnetic tape, non-volatile memory cards, and ROMs.

[0073] Furthermore, the program code that realizes the functions described in this embodiment can be implemented in a wide range of programming or scripting languages, such as assembler, C / C++, perl, Shell, PHP, Python, and Java.

[0074] Furthermore, the program code of the software that realizes the functions of the embodiments may be distributed via a network and stored in a storage means such as a computer's hard disk or memory, or in a storage medium such as a CD-RW or CD-R, and the processor of the computer may read and execute the program code stored in the storage means or the storage medium.

[0075] In the above-described embodiment, the control lines and information lines are shown as those considered necessary for the explanation, and not all control lines and information lines are necessarily shown in the product. All components may be interconnected.

Claims

1. A computer system comprising: a processor, a storage device connected to the processor, and a network interface connected to the processor; and accessible connection to a database that stores the read results of the base sequences of a plurality of biopolymers and the decay rates corresponding to the biopolymers; the processor: for each of the biopolymers, generates data representing a depth, which is the number of overlaps of the read results at each position of the base sequence of the biopolymer; for each of the biopolymers, calculates a standard deviation of the depth of the biopolymer based on the data; classifies the plurality of biopolymers into a plurality of groups based on the standard deviation of the depth of each of the biopolymers; for each of the groups, generates training data consisting of the base sequences of the biopolymers belonging to the group and the decay rates stored in the database; and for each of the groups, performs machine learning using the training data of the group to generate a prediction model that accepts the base sequence of the biopolymer and outputs the decay rate of the biopolymer, and stores information about the prediction model in the storage device.

2. A computer system according to claim 1, characterized in that the processor divides the base sequence of each of the biopolymers into a plurality of regions, calculates the decay rate of each of the regions of the biopolymer based on the data, and stores the calculated decay rate in the database.

3. A computer system as described in claim 2, wherein the database stores read results for the same biopolymer but with different elapsed times from the measurement reference time, and the processor uses the data for the same biopolymer with different elapsed times to calculate, for each region, the amount of change in depth relative to the change in the biopolymer over time, and calculates, for each region, the rate of decay of the biopolymer based on the amount of change in depth relative to the change in the biopolymer over time.

4. A computer system according to claim 1, characterized in that the processor classifies the biopolymers into a plurality of groups based on the standard deviation of the depth of each of the biopolymers and the number of the biopolymers.

5. A computer system according to claim 1, characterized in that the processor performs deep learning for each of the groups using the learning data of the group.

6. A computer system comprising a processor, a storage device connected to the processor, and a network interface connected to the processor, wherein the storage device stores information on a plurality of prediction models that receive the base sequence of a biopolymer as input and output the decay rate of the biopolymer, and the prediction model is a machine learning model that outputs the decay rate of the biopolymer, and is generated by the following processes: classifying a plurality of biopolymers into a plurality of groups based on the standard deviation of the depth of the biopolymer, which is calculated based on data representing the depth, which is the overlap number of read results at each position of the base sequence of the biopolymer; calculating the decay rate for each of the biopolymers; and performing machine learning for each of the groups using learning data consisting of the base sequences and decay rates of the biopolymers belonging to the group, and wherein the processor predicts the decay rate of the biopolymer to be predicted using the plurality of prediction models.

7. A computer system according to claim 6, wherein the processor, when receiving the base sequence of the biopolymer to be predicted, inputs the base sequence of the biopolymer to be predicted into each of the plurality of prediction models, calculates statistical values ​​of the decay rate of the biopolymer to be predicted based on the decay rates obtained from each of the plurality of prediction models, and selects the decay rate to be output based on the statistical values.

8. A computer system according to claim 6, wherein the storage device stores classification criteria for the groups, and the processor, when receiving the base sequence of the biopolymer to be predicted and the reading results, calculates the standard deviation of the depth of the biopolymer to be predicted based on the reading results, identifies the group to which the biopolymer to be predicted belongs based on the classification criteria and the standard deviation of the depth of the biopolymer to be predicted, and inputs the base sequence of the biopolymer to be predicted into the prediction model of the identified group.

9. A method for training a predictive model executed by a computer system, wherein the computer system has a processor, a storage device connected to the processor, and a network interface connected to the processor, and is accessible to a database that stores read results of the base sequences of a plurality of biopolymers and decay rates corresponding to the biopolymers, the method for training a predictive model comprising: a first step in which the processor generates data for each of the biopolymers that represents a depth, which is the number of overlaps of the read results at each position of the base sequence of the biopolymer; a second step in which the processor calculates, for each of the biopolymers, a standard deviation of the depth of the biopolymer based on the data; a third step in which the processor classifies the biopolymers into a plurality of groups based on the standard deviation of the depth of each of the biopolymers; a fourth step in which the processor generates training data for each of the groups, which is composed of the base sequences of the biopolymers belonging to the group and the decay rates stored in the database; and a fifth step in which the processor generates a predictive model for each of the groups and stores information of the predictive model in the storage device.

10. A method for training a prediction model as described in claim 9, characterized in that it includes a sixth step in which the processor divides the base sequence of each of the biopolymers into a plurality of regions, calculates the decay rate of each of the regions of the biopolymer based on the data, and stores the calculated rate in the database.

11. A method for training a prediction model as described in claim 9, characterized in that the third step includes a step in which the processor classifies the biopolymers into multiple groups based on the standard deviation of the depth of each of the biopolymers and the number of the biopolymers.

12. A method for training a predictive model as described in claim 9, wherein the fifth step comprises the processor performing deep learning for each of the groups using the training data for the group.

13. A method for training a prediction model according to claim 9, characterized in that the processor uses the plurality of prediction models to predict the decay rate of a biopolymer to be predicted.

14. A method for training a prediction model as described in claim 13, comprising the steps of: when receiving a base sequence of the biopolymer to be predicted, the processor inputs the base sequence of the biopolymer to be predicted into each of the plurality of prediction models; the processor calculates statistical values ​​of the decay rate of the biopolymer to be predicted based on the decay rate for each region obtained from each of the plurality of prediction models; and the processor outputs the statistical values.

Images