KR20230164808A - Transcriptome-based synthetic lethality prediction device, method and computer program - Google Patents

Abstract

According to embodiments of the present disclosure, a synthetic lethality prediction device includes the steps of: receiving data on expression changes for genes from an external database; determining a first number of input values suitable for predicting a synthetic lethal relationship using a feature selection unit; Obtaining a set of input values corresponding to the first number of expression changes for the genes; Inputting a set of expression changes of the first number of genes into at least one of first to eighth synthetic lethality prediction units, and learning and generating the first to eighth synthetic lethality prediction units; And inputting the expression changes for the genes into at least one of the first to eighth synthetic lethal prediction units, and outputting one or more genes in a new synthetic lethal relationship with the target gene. A method for predicting synthetic lethality, comprising: commences.

Description

Transcriptome data-based synthetic lethality prediction device, synthetic lethality prediction method, and computer program {Transcriptome-based synthetic lethality prediction device, method and computer program}

According to an embodiment of the present disclosure, it relates to a synthetic lethality prediction device based on transcriptome data, a synthetic lethality prediction method, and a computer program, and more specifically, selection of learned features based on data such as expression changes of genes for cell lines. It is characterized by inputting data about genes into the negative and synthetic lethality prediction parts and outputting whether there is a synthetic lethality relationship.

Cancer is a leading cause of death. A novel approach for developing treatments for cancer concerns the concept of synthetic lethality. If only one of the two genes that play complementary functions important for cancer cell survival is mutated, the cancer cell can survive, but if both genes are mutated, the cancer cell dies, which is called synthetic lethality. In other words, synthetic lethality describes a situation in which a mutation and a drug work together to inhibit the function of two genes, causing the death of cancer cells (-neither the mutation nor the drug causes the death of cancer cells). Targeting a synthetic lethal gene (or gene product) to a cancer-related mutation kills only cancer cells and allows normal cells to survive. Therefore, synthetic lethality provides a framework for the development of anti-cancer specific agents.

Therefore, in order to develop anticancer treatments related to target genes, there is a need to discover new genes that are related to synthetic lethality with the target gene.

The purpose of the embodiments disclosed in this specification is to present a synthetic lethality prediction device and method for detecting new synthetic lethality in addition to the previously discovered synthetic lethality.

The purpose of the embodiments disclosed herein is to present a synthetic lethality prediction device and method that can be used to discover targets and biomarkers for new drug development using novel synthetic lethality.

Embodiments disclosed in this specification detect new synthetic lethality for each gene, use this novel synthetic lethality to discover targets for new drug development, and increase the probability of success in new drug development through biomarker-based clinical development. The purpose is to present a synthetic fatality prediction device and method to increase .

The method according to embodiments of the present disclosure includes the steps of: a synthetic lethality prediction device receiving data on expression changes in genes from an external database; determining a first number of input values suitable for predicting a synthetic lethal relationship using a feature selection unit; Obtaining a set of input values corresponding to the first number of expression changes for the genes; Inputting a set of expression changes of the first number of genes into at least one of first to eighth synthetic lethality prediction units, and learning and generating the first to eighth synthetic lethality prediction units; and inputting the expression changes for the genes into at least one of the first to eighth synthetic lethal prediction units, and outputting one or more genes in a new synthetic lethal relationship with the target gene.

The first to eighth synthetic lethal prediction units each include a plurality of prediction models trained with genes from one of the first to eighth cell lines, and output values from the plurality of prediction models are combined to produce a final output. The value can be output.

The feature selection unit may input the expression change amount of a plurality of genes in one of the first to eighth cell lines and output the number of input values suitable for predicting a synthetic lethal relationship as an output value.

The plurality of prediction models included in the first to eighth synthetic lethality prediction units may be models that are learned using expression changes in the remaining genes after excluding one of the genes highly related to synthetic lethality stored as input values.

A plurality of prediction models included in the first to eighth synthetic lethal prediction units may be generated as many as the number of genes highly related to the synthetic lethality.

The feature selection unit determines the expression change amount of genes in the cell line by one of undersampling or oversampling according to the result of comparing the number of genes with synthetic lethality and the number of genes with non-synthetic lethality contained in the cell line. It can be learned by processing data about .

The synthetic fatality prediction device according to embodiments of the present disclosure includes a processor, a communication unit, and a memory, and the processor executes instructions stored in the memory,

Receive data on expression conversion amounts for genes from an external database, determine a first number of input values suitable for predicting a synthetic lethal relationship using a feature selection unit, and determine expression values of the target gene and the remaining genes. Among them, a set of input values equal to the first number is obtained, and a set of expression changes of the first number of genes is input to at least one of the first to eighth synthetic lethality prediction units, and the first to eighth synthesis Lethal prediction units are learned and generated, and expression changes for the genes are input into at least one of the first to eighth synthetic lethal prediction units, and one or more genes in a new synthetic lethal relationship with the target gene can be output. .

A computer program according to an embodiment of the present invention may be stored in a medium to execute any one of the methods according to an embodiment of the present invention using a computer.

In addition to this, another method for implementing the present invention, another system, and a computer-readable recording medium for recording a computer program for executing the method are further provided.

Other aspects, features and advantages in addition to those described above will become apparent from the following drawings, claims and detailed description of the invention.

According to any one of the means for solving the above-mentioned problem, it is possible to detect new synthetic lethality other than the previously discovered synthetic lethality.

According to any one of the above-described means of solving the problem, novel synthetic lethality can be used to discover targets and biomarkers for new drug development.

According to one of the above-mentioned problem solving methods, novel synthetic lethality is detected for each gene, these novel synthetic lethality is used to discover targets for new drug development, and new drug development through biomarker-based clinical development is performed. You can increase your chances of success.

Figure 1 is a block diagram of a synthetic fatality prediction device 100 according to embodiments of the present disclosure.
Figure 2 illustrates a synthetic fatality prediction network system including a server and a user terminal according to an embodiment of the present disclosure.
Figure 3 is a detailed block diagram of the feature selection unit 120 and the synthetic fatality prediction unit 130 according to an embodiment of the present disclosure.
Figure 4 is a block diagram of the first sub synthetic fatality prediction unit 131.
Figure 5 is a block diagram of the feature selection learning device 200 that trains the feature selection unit.
Figure 6a is an example diagram of a model evaluation table for cell line A549 according to an embodiment of the present disclosure.
Figure 6b is an example diagram of an evaluation table of a model for cell line SLDB according to an embodiment of the present disclosure.
Figure 7 is a graph of F1 score values according to the number of features in the feature selection unit.
Figure 8 is a graph of F1 score values according to the number of features of the feature selection unit according to another embodiment.
Figure 9 is a block diagram of the prediction unit learning device 300 that trains the synthetic fatality prediction unit.
FIG. 10 is an example diagram of a network environment including the synthetic fatality prediction device 100 according to embodiments of the present disclosure.
Figure 11 is a flowchart of a synthetic fatality prediction method according to embodiments of the present disclosure.

Hereinafter, the configuration and operation of the present invention will be described in detail with reference to embodiments of the present invention shown in the attached drawings.

Since the present invention can be modified in various ways and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail in the detailed description. The effects and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various forms.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. When describing with reference to the drawings, identical or corresponding components will be assigned the same reference numerals and redundant description thereof will be omitted. .

In this specification, terms such as “learning” and “learning” are not intended to refer to mental operations such as human educational activities, but are terms that refer to performing machine learning through procedural computing. interpret.

In the following embodiments, terms such as first and second are used not in a limiting sense but for the purpose of distinguishing one component from another component.

In the following examples, singular terms include plural terms unless the context clearly dictates otherwise.

In the following embodiments, terms such as include or have mean the presence of features or components described in the specification, and do not preclude the possibility of adding one or more other features or components.

In the drawings, the sizes of components may be exaggerated or reduced for convenience of explanation. For example, the size and thickness of each component shown in the drawings are shown arbitrarily for convenience of explanation, so the present invention is not necessarily limited to what is shown.

In cases where an embodiment can be implemented differently, a specific process sequence may be performed differently from the described sequence. For example, two processes described in succession may be performed substantially at the same time, or may be performed in an order opposite to that in which they are described.

In the present disclosure below, the meaning that the server controls the user terminal means that the server outputs output from the user terminal through communication with the user terminal (all output devices from the user terminal, such as screen display, sound output, vibration output, lamp emission, etc.) And it may mean that the user terminal provides data for performing a certain operation. Of course, the server may control the output from the user terminal using data previously stored in the user terminal, and is not limited to the above example.

In the present disclosure below, the meaning of transmitting and receiving information or data with a user account may include the meaning of transmitting and receiving information or data with a device (or user terminal) corresponding to or linked to the user account.

In the present disclosure below, the user account corresponding to the user terminal may include a user account that logs in or accesses the service through the user terminal and a user account in which the user terminal stores information. Additionally, the user terminal of the user account may mean the user terminal to which the user account is logged in, user account information is stored, or the user account is connected.

In the present disclosure below, the synthetic fatality prediction device may be a user terminal, a server, a synthetic fatality prediction system, or a separate device.

In the present disclosure below, synthetic lethality refers to a phenomenon in which cell death occurs when two genes are simultaneously knocked out, knocked down, or perturbed by drugs.

In the present disclosure below, a cell line may refer to a set of cells that can be continuously cultured outside the body.

According to the present disclosure below, it is possible to detect novel synthetic lethality other than the previously discovered synthetic lethality.

According to the present disclosure below, novel synthetic lethality can be used to discover anti-cancer targets for new drug development.

According to the present disclosure below, novel synthetic lethality is detected for each gene, and these new synthetic lethality are used to discover targets for new drug development and increase the probability of success in new drug development through biomarker-based clinical development. You can.

According to embodiments of the present disclosure, a synthetic lethality prediction device can newly predict a gene and a gene in a synthetic lethal relationship using a model learned based on expression changes of genes.

Figure 1 is a block diagram of a synthetic fatality prediction device 100 according to embodiments of the present disclosure.

The synthetic lethality prediction device 100 can input data on genes and output a set of genes with a synthetic lethality relationship newly predicted from the genes.

The synthetic lethality prediction device 100 may include a data input unit 110 that inputs data on expression changes for genes.

The synthetic lethality prediction device 100 includes a feature selection unit 120 and can output data required to predict a new gene set with a synthetic lethality relationship. The feature selection unit 120 may be implemented by learning based on the gene expression level for the cell line. For example, the feature selection unit 120 may be learned using a total of eight cancer cell lines: A375, A549, HA1E, HCC515, HEPG2, HT29, MCF7, and PC3, and may include eight feature selection units. It is not limited to eight cancer cell lines and can be learned and implemented based on various cell lines.

The feature selection unit 120 may be trained by inputting data including expression changes of genes. The feature selection unit 120 may be supervised using a tree-based feature selection model, but is not limited to this. More specifically, the feature selection unit 120 may select n genes extracted based on the feature importance of the random forest model. Here, n is a natural number. In addition, the feature selection unit 120 can select n genes using methods such as lightGBM, KNN, Logistic regression, Naive Bayes, and SVM. Finally, n gene sets can be output. n and n gene sets may be determined differently for each cell line. The feature selection unit 120 can be learned by inputting expression changes of genes and a set of genes with a synthetic lethal relationship according to a supervised learning method. Here, data on changes in expression of genes may refer to data based on the results of drug perturbation or gene interference experiments performed on genes of target cells. Data on expression changes of genes may be obtained by acquiring raw data, converting the data into an image, quantile normalization results for landmark genes, and inferred data for genes other than landmark genes, in the form of an array, Or it may be in the form of a 2X2 matrix. Here, the quantile normalization result refers to the resulting data generated after going through a correction process for the data generated as an experiment result. Data generated as a result of an experiment may include deviations depending on the testing environment, test performance organization, etc. The quantile normalization result can refer to the data resulting from correcting this deviation.

The synthetic lethality prediction device 100 includes a synthetic lethality prediction unit 130 and can determine whether the genes have a synthetic lethality relationship through a supervised learning prediction model by inputting a set of expression changes of genes.

The synthetic lethality prediction unit 130 may be implemented as a result of supervised learning for each cell line by acquiring gene expression change data for each cell line.

Even for one cell line, multiple learning models can be implemented and results from multiple learning models can be combined to output (return) whether or not there is a synthetic lethal relationship. Results from multiple learning models may include whether there is a synthetic lethal relationship (Y, N).

At this time, a plurality of learning models learned for one cell line may be learned by excluding genes depending on the relationship with the disease, matching with the synthetic lethal relationship, etc. For example, the synthetic lethality prediction unit 130 acquires data on 76 genes that are synthetic lethal and are related to cancer, and supervised learning a learning model with a learning data set in which one gene among the 76 genes is excluded. You can. Since each of the 76 genes is excluded and learned, multiple learning models can be created as many as 76. The synthetic lethal prediction unit 130 may generate a plurality of learning models learned by excluding each of the genes related to cancer and having a synthetic lethal relationship for each cell line. The synthetic lethality prediction unit 130 may output the result that occurred with the greatest frequency among the results of the models learned with the first cell line as the first result value of the model learned with the first cell line. The synthetic lethal prediction unit 130 may input models learned using the first to eighth cell lines and output first to eighth result values.

The determination unit 140 may determine whether there is a synthetic lethal relationship between genes by using the first to eighth result values as input.

The learning model included in the feature selection unit 120 and/or the learning model included in the synthetic lethality prediction unit 130 is based on 1574 genes, including 978 landmark genes and 596 DNA repair genes implemented in the LINCS L1000 project. It can be learned from data about The data input to the learning model may include the results of about 3,000 genetic interference experiments, but is not limited to this and various modifications may be possible. Data input to the learning model may include quantile normalization results for the landmark gene and inference data for genes other than the landmark gene.

The learning model included in the feature selection unit 120 or the learning model included in the synthetic lethal prediction unit 130 is generated for known cell lines such as A375, A549, HA1E, HCC515, HEPG2, HT29, MCF7, PC3, etc., Each cell line can be generated separately.

For example, a learning model related to the A549 cell line can be trained with 126 gene pairs known to be synthetic lethal and 1094 gene pairs that are not synthetic lethal as input.

The learning models related to the remaining seven cancer cell lines, A375, HA1E, HCC515, HEPG2, HT29, MCF7, and PC3, can be trained using 1,164 synthetic lethal gene pairs and 416 gene pairs that do not cause synthetic lethality as input.

Each learning model can be trained using an undersampling method. In other words, the learning model related to the A549 cell line can be trained using 126 gene pairs as the correct answer. Learning models related to the remaining seven cancer cell lines, A375, HA1E, HCC515, HEPG2, HT29, MCF7, and PC3, can be trained using 416 gene pairs as the answer sheet.

The learning model included in the feature selection unit 120 and/or the learning model included in the synthetic fatality prediction unit 130 may utilize the lightGBM algorithm or Neural Network algorithm.

According to an embodiment of the present disclosure, synthetic lethality can be predicted based on the amount of expression change for the gene selected in the feature selection unit 120. Through this, the synthetic lethal relationship of the input genes can be predicted even through training data in which the number of genes with a synthetic lethal relationship is different from the number of genes with a non-synthetic lethal relationship.

Figure 2 illustrates a synthetic fatality prediction network system including a server and a user terminal according to an embodiment of the present disclosure.

The synthetic fatality prediction network system 1 of the present disclosure may include a server 20 and at least one user terminal 11 to 16. The server 20 can provide services for analyzing various genes through a network. The server 20 may provide a service for analyzing genes to at least one user terminal 11 to 16 at the same time.

According to an embodiment of the present disclosure, the server 20 may include a single server, a collection of servers, a cloud server, etc., and is not limited to the above examples. The server 20 may include a database that stores data for analyzing genes, data on genetic mutations, data on dependency scores, data on cell lines, data on expression levels, etc. As previously described, the server 20 may be a synthetic fatality prediction device.

According to an embodiment of the present disclosure, a network refers to a connection established (or formed) using all communication methods, and refers to a communication network connected through all communication methods that transmits and receives data between terminals and terminals or between terminals and servers. can do.

All communication methods may include all communication methods, such as communication through a certain communication standard, a certain frequency band, a certain protocol, or a certain channel. For example, it may include Bluetooth, BLE, Wi-Fi, Zigbee, 3G, LTE, and ultrasonic communication methods, and may include short-range communication, long-distance communication, wireless communication, and wired communication. Of course, it is not limited to the above example.

According to an embodiment of the present disclosure, the short-distance communication method may mean a communication method that allows communication only when the device (terminal or server) performing communication is within a predetermined range, for example, Bluetooth, It may include NFC, etc. A long-distance communication method may refer to a communication method in which a device performing communication can communicate regardless of the distance. For example, a long-distance communication method may refer to a method in which two devices that communicate through a repeater such as an AP can communicate even when they are over a predetermined distance, and can use cellular networks (3G, LTE) such as SMS and phone calls. It may include the communication method used. Of course, it is not limited to the above example. Receiving a genetic analysis service using a network may mean that communication between the server and the terminal can be performed through any communication method.

Throughout the specification, at least one user terminal (11 to 16) refers to a personal computer (11), a tablet (12), a cellular phone (13), a laptop (14), and a smart phone. (15), In addition to the TV (16), it may include various electronic devices such as PDAs (Personal Digital Assistants), PMPs (Portable Multimedia Players), navigation, MP3 players, digital cameras, refrigerators, washing machines, and vacuum cleaners. It is not limited to examples. As described above, at least one user terminal 11 to 16 may be a synthetic fatality prediction device.

According to an embodiment of the present disclosure, the genetic analysis service includes a service providing a synthetic lethal candidate gene group related to a target gene, a service providing cell line data related to the target gene, and a service providing whether synthetic lethality exists between the target gene and the candidate gene. It may include services, services that provide the degree of similarity between target genes and candidate genes, etc., but is not limited to the examples above.

According to an embodiment of the present disclosure, the server 20 may determine whether a pair of genes may have a synthetic lethal relationship by inputting data according to the amount of change in gene expression. The server 20 uses a feature selection unit learned by inputting the amount of gene expression change for each cell line, and a synthetic lethality prediction unit learned by inputting the change in expression of genes equal to the number of features in the feature selection unit to determine whether genes are synthetically lethal. can be judged. At this time, the synthetic lethality prediction unit may include a plurality of prediction models learned by excluding one of the designated genes. The genes determined here may be determined by their relationship with diseases such as cancer or their relationship with synthetic lethality. Data on designated genes can be stored in advance or obtained by requesting an external database. A plurality of prediction models may be generated by alternately excluding one of the designated genes. For example, if there are 10 designated genes, 10 multiple prediction models can be created by excluding each of the 10 genes. The server 20 may determine the most frequent result among the results from the plurality of prediction models as the final result.

According to an embodiment of the present disclosure, one of the at least one user terminal (11 to 16) can determine whether a pair of genes can be in a synthetic lethal relationship by inputting data according to the amount of change in gene expression. . One of the at least one user terminal (11 to 16) is a feature selection unit learned by inputting the change in gene expression for each cell line, and a synthetic lethality prediction learned by inputting the change in expression of genes corresponding to the number of features in the feature selection unit. Using the wealth, it is possible to determine whether genes are synthetically lethal. At this time, the synthetic lethality prediction unit may include a plurality of prediction models learned by excluding one of the designated genes. The genes determined here may be determined by their relationship with diseases such as cancer or their relationship with synthetic lethality. Data on designated genes can be stored in advance or obtained by requesting an external database. A plurality of prediction models may be generated by alternately excluding one of the designated genes. For example, if there are 10 designated genes, 10 multiple prediction models can be created by excluding each of the 10 genes. One of the at least one user terminal 11 to 16 may determine the most frequent result among the results from the plurality of prediction models as the final result.

Additionally, according to an embodiment of the present disclosure, the synthetic lethality prediction network system 1 can determine whether a pair of genes can have a synthetic lethal relationship by inputting data according to the amount of change in gene expression. The synthetic lethality prediction network system (1) uses a feature selection unit learned by inputting the change in gene expression for each cell line, and a synthetic lethality prediction unit learned by inputting the change in expression of genes equal to the number of features in the feature selection unit. It is possible to determine whether or not they are synthetically lethal. At this time, the synthetic lethality prediction unit may include a plurality of prediction models learned by excluding one of the designated genes. The genes determined here may be determined by their relationship with diseases such as cancer or their relationship with synthetic lethality. Data on designated genes can be stored in advance or obtained by requesting an external database. A plurality of prediction models may be generated by alternately excluding one of the designated genes. For example, if there are 10 designated genes, 10 multiple prediction models can be created by excluding each of the 10 genes. The synthetic fatality prediction network system 1 may determine the most frequent result among the results from a plurality of prediction models as the final result.

Figure 3 is a detailed block diagram of the feature selection unit 120 and the synthetic fatality prediction unit 130 according to an embodiment of the present disclosure.

The synthetic lethal prediction device 100 according to embodiments of the present disclosure can be implemented as shown in FIG. 3 by learning about eight cell lines. In Figure 3, the first to eighth sub-feature selection units (121, 122, 123, 124, 125, 126, 127, 128) are shown to include eight sub-feature selection units, but are not limited to this and include various sub-feature selection units. It may be implemented to include a number of sub-feature selection units.

The first sub-feature selection unit 121 is learned from data on the amount of gene expression change in the A375 cell line, and can output the number of genes used for predicting synthetic lethality from the data on the A375 cell line as a result of learning.

The second sub-feature selection unit 122 is learned from data on the amount of gene expression change in the A549 cell line, and can output the number of genes used for predicting synthetic lethality from the data on the A549 cell line as a result of learning.

The third sub-feature selection unit 123 is learned from data on the amount of gene expression change in the HA1E cell line, and can output the number of genes used for predicting synthetic lethality from the data on the HA1E cell line as a result of learning.

The fourth sub-feature selection unit 124 is learned from data on the amount of gene expression change in the HCC515 cell line, and can output the number of genes used for predicting synthetic lethality from the data on the HCC515 cell line as a result of learning.

The fifth sub-feature selection unit 125 is learned with data on the amount of gene expression change in the HEPG2 cell line, and can output the number of genes used for predicting synthetic lethality from the data on the HEPG2 cell line as a result of learning.

The sixth sub-feature selection unit 126 is learned from data on the amount of gene expression change in the HT29 cell line, and as a result of learning, can output the number of genes used for predicting synthetic lethality from the data on the HT29 cell line.

The seventh sub-feature selection unit 127 is learned from data on the amount of gene expression change in the MCF7 cell line, and can output the number of genes used for predicting synthetic lethality from the data on the MCF7 cell line as a result of learning.

The eighth sub-feature selection unit 128 is learned with data on the amount of gene expression change in the PC3 cell line, and can output the number of genes used for predicting synthetic lethality from the data on the PC3 cell line as a result of learning.

The feature selection unit 120 may calculate output values obtained from the first to eighth sub-feature selection units 128, respectively, in order to determine whether there is a synthetic lethal relationship with the input genes. The eight output values obtained from the first to eighth sub-feature selection units 128 may be transmitted to the synthetic fatality prediction unit 130 and used to predict the synthetic lethality relationship.

The synthetic lethal prediction unit 130 includes the first to eighth sub synthetic lethal prediction units (131, 132, 133, …, 138) learned for a total of eight cell lines: A375, A549, HA1E, HCC515, HEPG2, HT29, MCF7, and PC3. ) can be used to output the synthetic lethality relationship of the input genes.

The first sub-synthetic lethality prediction unit 131 is learned using the gene expression change obtained through the A375 cell line as input, and selects genes used for learning based on the results of the first sub-feature selection unit 121. You can. When the first sub synthetic lethality prediction unit 131 receives the amount of gene expression change for the A375 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the first sub feature selection unit 121. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The second sub-synthetic lethality prediction unit 132 is learned using the gene expression change obtained through the A549 cell line as input, and selects genes used for learning based on the results of the second sub-feature selection unit 122. You can. When the second sub synthetic lethality prediction unit 132 receives the amount of gene expression change for the A549 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the second sub feature selection unit 122. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The third sub-synthetic lethality prediction unit 133 is learned using the gene expression change obtained through the HA1E cell line as input, and selects genes used for learning based on the results of the third sub-feature selection unit 123. You can. When the third sub synthetic lethality prediction unit 133 receives the amount of gene expression change for the HA1E cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the third sub feature selection unit 123. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The fourth sub-synthetic lethality prediction unit 134 is learned using the gene expression change obtained through the HCC515 cell line as input, and selects genes used for learning based on the results of the fourth sub-feature selection unit 124. You can. When the fourth sub synthetic lethality prediction unit 134 receives the amount of gene expression change for the HCC515 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the fourth sub feature selection unit 124. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The fifth sub synthetic lethality prediction unit 135 is learned using the gene expression change obtained through the HEPG2 cell line as input, and selects genes used for learning based on the results of the fifth sub feature selection unit 125. You can. When the fifth sub synthetic lethality prediction unit 135 receives the amount of gene expression change for the HEPG2 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the fifth sub feature selection unit 125. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The sixth sub-synthetic lethal prediction unit 136 is learned using the gene expression change obtained through the HT29 cell line as input, and selects genes used for learning based on the results of the sixth sub-feature selection unit 126. You can. When the sixth sub synthetic lethality prediction unit 136 receives the amount of gene expression change for the HT29 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the sixth sub feature selection unit 126. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The 7th sub synthetic lethality prediction unit 137 is learned using the gene expression change obtained through the MCF7 cell line as input, and selects genes used for learning based on the results of the 7th sub feature selection unit 127. You can. When the 7th sub synthetic lethality prediction unit 137 receives the gene expression change amount for the MCF7 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the 7th sub feature selection unit 127. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The 8th sub-synthetic lethality prediction unit 138 is learned using the gene expression change obtained through the PC3 cell line as input, and selects genes used for learning based on the results of the 8th sub-feature selection unit 128. You can. When the 8th sub synthetic lethality prediction unit 138 receives the amount of gene expression change for the PC3 cell line, it corresponds to the result value (set of selection genes, number of selection genes, etc.) of the 8th sub feature selection unit 128. The synthetic lethal relationship of genes can be output by using the change in expression of the gene as input.

The synthetic lethal prediction unit 130 may determine whether the input gene is synthetic lethal by combining the results output from the first to eighth sub synthetic lethal prediction units 131, 132, 133, ..., 138.

Figure 4 is a block diagram of the first sub synthetic fatality prediction unit 131.

As shown in FIG. 4, the first sub synthetic fatality prediction unit 131 includes a 1-1s sub synthetic fatality prediction unit 131S1, a 1-2s sub synthetic fatality prediction unit 131S2,... , It may be implemented including a 1-76s sub synthetic fatality prediction unit 131S76 and a first prediction value determination unit 131-o. Similar to the structure of the first sub-synthetic fatality prediction unit 131, the second to eighth sub-synthetic fatality prediction units may also be implemented including 76 sub synthetic lethality prediction units and a prediction value determination unit.

The first sub-synthetic lethal prediction unit 131 is a sub-synthetic lethal prediction unit that excludes some of the genes that are synthetic lethal and highly related to the disease in order to obtain independent results that exclude the synthetic lethal relationship and the relationship with the disease. can be learned. Genes that are synthetic lethal and highly related to disease are stored in advance. Figure 4 shows an example of 76 genes implemented.

The first sub synthetic lethality prediction unit 131 can learn the 1-1s sub synthetic lethality prediction unit 131S1 based on gene expression change data excluding the first gene among 76 genes. Sequentially, the first sub-synthetic lethality prediction unit 131 trains the 1-ks sub-synthetic lethality prediction unit 131Sk based on gene expression change data excluding the k-th gene among the 76 genes, thereby The through 76s sub synthetic fatality prediction units can be generated. Here, k refers to a natural number between 1 and 76.

As a result, the gene sets input to the 1-1s to 1-76s sub synthetic lethal prediction units (131S1, 131S2, 131S3, ... 131S76) partially match but do not completely match. By using 76 sub-synthetic lethal prediction units trained with this completely mismatched gene set, a model with high independence from genes highly related to synthetic lethality can be created.

The first predicted value determination unit 131-o determines the result value that occurs with the highest probability among the result values from the 1-1s to 1-76s sub synthetic fatality prediction units 131S1, 131S2, 131S3, ... 131S76. It can be output as the final result.

Figure 5 is a block diagram of the feature selection learning device 200 that trains the feature selection unit.

The feature selection learning device 200 can generate a feature selection unit and a sub-feature selection unit with a predetermined quality by learning the feature selection unit and the sub-feature selection unit.

The data input unit 210 performs the function of inputting data about genes required for learning. The data input unit 210 can input data such as the amount of change in gene expression for cell lines.

The data input to the data input unit 210 is based on the results of over 3,000 drug perturbation and gene interference experiments on 1,574 genes including 978 landmark genes and 596 DNA repair genes conducted in the LINCS L1000 project. Results from cell lines may be included. Quantile normalization results for the landmark gene and inference data for genes other than the landmark gene may be input into the data input unit 210.

The model learning unit 220 can train the feature selection unit using the input data. The model learning unit 220 can learn different feature selection units for each cell line. The model learning unit 220 may train one or more feature selection units. The model learning unit 320 may process the input data using methods such as under-sampling and over-sampling and train the feature selection unit. The model learning unit 220 can be created by learning one or more feature selection units using methods such as lightGBM, KNN, Logistic regression, Naive bayes, Random forest SVM, and EXP2SL.

The model evaluation unit 230 may evaluate the learned feature selection units to rescue a feature selection unit with good performance. As shown in FIG. 6A, the model evaluation unit 230 calculates the accuracy, area under the ROC curve (AUROC), Matthews correlation coefficient (MCC), and F1 of the feature selection units learned by inputting the cell line of A549. By comparing scores, etc., you can determine the model with the best performance. In the model evaluation unit 230, a feature selection unit generated by the LightGBM and EXP2SL methods can be finally constructed. The performance of the feature selection unit may vary depending on the cell line of the input data. As shown in Figure 6b, when trained with the SLDB data set, the feature selection unit generated by the LightGBM method appears to have the best performance.

Figure 7 is a graph of F1 score values according to the number of features in the feature selection unit.

The feature selection learning device 200 can generate a feature selection unit that can predict a synthetic lethal relationship with a set of 20 genes as a result of learning with data on the cell line A549.

Figure 8 is a graph of F1 score values according to the number of features of the feature selection unit according to another embodiment.

The feature selection learning device 200 can generate a feature selection unit capable of predicting a synthetic lethal relationship with a set of 60 genes as a result of learning with data on other cell lines.

Figure 9 is a block diagram of the prediction unit learning device 300 that trains the synthetic fatality prediction unit.

The prediction unit learning device 300 can generate a synthetic lethality prediction unit and a sub-synthetic lethality prediction unit with a predetermined accuracy or higher by training the synthetic lethality prediction unit and the sub-synthetic lethality prediction unit.

The data input unit 310 performs the function of inputting data about genes required for learning. The data input unit 310 can input data such as the amount of change in gene expression for cell lines.

The data input to the data input unit 310 is based on the results of over 3,000 drug perturbation and gene interference experiments on 1,574 genes including 978 landmark genes and 596 DNA repair genes conducted in the LINCS L1000 project. Results from cell lines may be included. Quantile normalization results for the landmark gene and inference data for genes other than the landmark gene may be input into the data input unit 310.

The model learning unit 320 can train the synthetic fatality prediction unit using the input data. The model learning unit 320 can learn different synthetic lethality prediction units for each cell line. The synthetic lethality prediction unit for each cell line can be learned as an ensemble model including a plurality of learning models. The model learning unit 320 may train one or more synthetic fatality prediction units. The model learning unit 320 may process the input data using methods such as under-sampling and over-sampling and train the synthetic fatality prediction unit. The model learning unit 320 can be generated by learning one or more synthetic fatality prediction units using methods such as lightGBM, KNN, Logistic regression, Naive Bayes, Random forest, SVM, and EXP2SL.

FIG. 10 is an example diagram of a network environment including the synthetic fatality prediction device 100 according to embodiments of the present disclosure.

At least two of the synthetic fatality prediction device 100, the feature selection learning device 200, the synthetic fatality prediction learning device 300, the database 400, and the user terminal 500 may be connected to a network and communicate.

The synthetic fatality prediction device 100 may receive a feature selection unit learned from the feature selection learning device 200 and input data into the feature selection unit. The synthetic lethality prediction device 100 may receive a synthetic lethality prediction unit learned from the synthetic lethality prediction learning device 300 and input data into the synthetic lethality prediction unit. The synthetic lethality prediction device 100 can determine and output new synthetic lethality relationships for genes using a feature selection unit and a synthetic lethality prediction unit. The data output in this way can be transmitted to the database 400 and stored. The feature selection learning device 200 can receive data such as newly input gene expression change amount data and whether there is a synthetic lethal relationship from the synthetic lethality prediction device 100 or the database 400, and perform learning again with the received data. there is. The synthetic lethality prediction learning device 300 receives newly input data such as gene expression change amount data and synthetic lethality relationship from the synthetic lethality prediction device 100 or the database 400 and performs learning again with the received data. You can. The synthetic fatality prediction device 100 may receive the feature selection unit and the synthetic fatality prediction unit generated as a result of this new learning and periodically update them.

The synthetic lethality prediction device 100 may periodically receive data on expression changes for genes input to the feature selection unit and the synthetic lethality prediction unit from the database 400 in order to predict a new synthetic lethality relationship.

Figure 11 is a flowchart of a synthetic fatality prediction method according to embodiments of the present disclosure.

In S110, the synthetic lethality prediction device can receive data on expression changes in genes from an external database.

In S120, the synthetic fatality prediction device may determine a first number of input values suitable for predicting a synthetic fatality relationship using a feature selection unit.

In S130, the synthetic lethality prediction device may obtain a set of input values corresponding to the first number of expression changes for genes.

In S140, the synthetic lethal prediction device may input a set of expression changes of a first number of genes into at least one of the first to eighth synthetic lethal prediction units, and learn and generate the first to eighth synthetic lethal prediction units. there is.

In S150, the synthetic lethal prediction device may input expression changes for genes into at least one of the first to eighth synthetic lethal prediction units and output one or more genes in a new synthetic lethal relationship with the target gene.

The device described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), It may be implemented using one or more general-purpose or special-purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. Additionally, a processing device may access, store, manipulate, process, and generate data in response to the execution of software. For ease of understanding, a single processing device may be described as being used; however, those skilled in the art will understand that a processing device includes multiple processing elements and/or multiple types of processing elements. It can be seen that it may include. For example, a processing device may include a plurality of processors or one processor and one controller. Additionally, other processing configurations, such as parallel processors, are possible.

Software may include a computer program, code, instructions, or a combination of one or more of these, which may configure a processing unit to operate as desired, or may be processed independently or collectively. You can command the device. Software and/or data may be used on any type of machine, component, physical device, virtual equipment, computer storage medium or device to be interpreted by or to provide instructions or data to a processing device. , or may be permanently or temporarily embodied in a transmitted signal wave. Software may be distributed over networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., singly or in combination. Program instructions recorded on the medium may be specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. -Includes optical media (magneto-optical media) and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, etc. Examples of program instructions include machine language code, such as that produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

As described above, although the embodiments have been described with limited examples and drawings, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims (13)

A synthetic lethal prediction device receiving data on expression changes in genes from an external database;
determining a first number of input values suitable for predicting a synthetic lethal relationship using a feature selection unit;
Obtaining a set of input values corresponding to the first number of expression changes for the genes;
Inputting a set of expression changes of the first number of genes into at least one of first to eighth synthetic lethality prediction units, and learning and generating the first to eighth synthetic lethality prediction units; and
Inputting the expression changes for the genes into at least one of the first to eighth synthetic lethal prediction units, and outputting one or more genes in a new synthetic lethal relationship with the target gene. A method for predicting synthetic lethality, comprising: According to paragraph 1,
The first to eighth synthetic fatality prediction units,
Containing a plurality of prediction models each trained with genes in one of the first to eighth cell lines,
A synthetic fatality prediction method that outputs a final output value by combining output values from a plurality of prediction models. According to paragraph 1,
The feature selection unit,
A synthetic lethality prediction method that inputs expression changes of a plurality of genes in one of the first to eighth cell lines and outputs the number of input values suitable for predicting a synthetic lethality relationship as an output value. According to paragraph 2,
The plurality of prediction models included in the first to eighth synthetic fatality prediction units are,
A synthetic lethality prediction method, which is a model that excludes one of the genes highly related to cancer development and learns the expression changes of the remaining genes as input. According to paragraph 4,
The plurality of prediction models included in the first to eighth synthetic fatality prediction units are,
A method for predicting synthetic lethality, which is generated as many genes as are highly related to the synthetic lethality. According to paragraph 1,
The feature selection unit,
Depending on the result of comparing the number of genes with synthetic lethality and the number of genes with non-synthetic lethality contained in the cell line, data on the expression changes of genes in the cell line are processed using either undersampling or oversampling. A synthetic fatality prediction method that is learned. Includes a processor, a communication unit, and a memory,
The processor executes instructions stored in the memory,
Receive data on expression conversion amounts for genes from an external database,
Determine a first number of input values suitable for predicting a synthetic lethal relationship using a feature selection unit,
Obtaining a set of input values equal to the first number among the expression values of the target gene and the remaining genes,
A set of expression changes of the first number of genes is input to at least one of the first to eighth synthetic lethality prediction units, and the first to eighth synthetic lethality prediction units are trained and generated,
A synthetic lethal prediction device that inputs the expression changes for the genes into at least one of the first to eighth synthetic lethal prediction units and outputs one or more genes in a new synthetic lethal relationship with the target gene. In clause 7,
The first to eighth synthetic fatality prediction units,
Containing a plurality of prediction models each trained with genes in one of the first to eighth cell lines,
A synthetic fatality prediction device that outputs a final output value by combining output values from a plurality of prediction models. In clause 7,
The feature selection unit,
A synthetic lethality prediction device that is trained to input the expression change amount of a plurality of genes in one of the first to eighth cell lines and output as an output the number of input values suitable for predicting a synthetic lethality relationship. According to clause 8,
The plurality of prediction models included in the first to eighth synthetic fatality prediction units are,
A synthetic lethality prediction device, which is a model that excludes one of the genes highly associated with stored synthetic lethality and learns the expression changes of the remaining genes as input. According to clause 10,
The plurality of prediction models included in the first to eighth synthetic fatality prediction units are,
A synthetic lethality prediction device that is generated as many genes as are highly related to the synthetic lethality. In clause 7,
The feature selection unit,
Depending on the result of comparing the number of genes with synthetic lethality and the number of genes with non-synthetic lethality contained in the cell line, data on the expression changes of genes in the cell line are processed using either undersampling or oversampling. A synthetic lethality prediction device that is learned. A computer program stored in a computer-readable storage medium to execute the method of claim 1 using a computer.

Images