KR102601304B1 - Method for diagnosis and therapeutic decision using artificial neural network trained with functional gene module and apparatus therefore - Google Patents

Abstract

A method and device for diagnosing and determining cancer treatment using a neural network that has learned gene function modules are disclosed. A method for estimating cancer and normal status using a neural network according to an embodiment of the present invention includes the steps of dividing the subject's genetic symbols into pathways, which are gene groups; Obtaining a signature for each of the divided pathways based on a neural network learned based on the pathway using each of the divided pathways as input; and estimating whether the pathway is cancerous or normal using the acquired signature.

Description

Method for diagnosis and therapeutic decision using artificial neural network trained with functional gene module and apparatus therefore}

The present invention relates to cancer diagnosis and treatment decision technology using a neural network that has learned gene function modules. More specifically, a deep neural network structured to enable life science interpretation and understanding is used to diagnose and treat cancer through deep learning. It relates to a method and device that can determine cancer/normal discrimination, cancer prognosis prediction, and customized cancer treatment using a learned deep neural network by learning the expression patterns of genes to distinguish normal cells.

The expression of genes determines the function of cells, and there are differences in gene expression in cancer and normal cells that contribute to differences in function. Even in the process of treating cancer, strategies to suppress or kill cancer by controlling the expression of specific genes are effective, and companion diagnostics that understand the expression patterns of specific genes and treat them with appropriate drugs are widely used clinically.

Existing companion diagnostics simply prescribe drugs through a combination of the expression of one or two or three genes. For example, Tratuzumab, which is widely used as a breast cancer treatment, is administered to breast cancer patients with overexpression of the HER2 gene, and as another example, Keytruda, a lung cancer immune checkpoint inhibitor, is used based on overexpression of the PD1 gene. Although the expression of a single gene meets clinical standards, it does not lead to ultimate cancer treatment, and efforts are continuing to understand the expression patterns of various genes and apply them to diagnosis and treatment.

However, because the number of genes is as many as 20,000 and their expression patterns are numerous, it is a challenge to discover whether specific expression patterns are related to cancer diagnosis, prognosis prediction, and treatment effectiveness.

Embodiments of the present invention use a learned deep neural network by learning the expression patterns of genes to distinguish cancer and normal cells through deep learning, a deep neural network structured to enable life science interpretation and understanding. Provides a method and device for determining normal cancer, predicting cancer prognosis, and determining customized treatment for cancer.

A method for estimating cancer and normal status using a neural network according to an embodiment of the present invention includes the steps of dividing the subject's genetic symbols into pathways, which are gene groups; Obtaining a signature for each of the divided pathways based on a neural network learned based on the pathway using each of the divided pathways as input; and estimating whether the pathway is cancerous or normal using the acquired signature.

The step of classifying the pathways is to calculate a pathway contribution index value for each of the pathways, and to select a preset number of upper pathway contribution index values among the calculated pathway contribution index values. After selecting genetic symbols having , the selected genetic symbols can be divided into the pathways.

In the step of acquiring the signature, a signature for each of the pathways may be acquired based on the sum of preset weights for each of the genetic symbols included in each of the pathways in the neural network.

The neural network may include pathway nodes connected one-to-one with each of the divided pathways.

Furthermore, the method for estimating cancer and normality using a neural network according to an embodiment of the present invention uses the signatures obtained for each of the pathways to determine the pathway and genetic symbols that contribute to estimating cancer and normality. An additional step of interpretation may be included.

A method for estimating cancer and normal status using a neural network according to another embodiment of the present invention includes the steps of selecting preset genetic symbols among the subject's genetic symbols; Classifying the selected genetic symbols into pathways, which are gene groups; Obtaining a signature for each of the divided pathways based on a neural network learned based on the pathway using each of the divided pathways as input; and estimating whether the pathway is cancerous or normal using the signature obtained for each of the pathways.

An apparatus for estimating cancer and normal status using a neural network according to an embodiment of the present invention includes a division unit that divides the subject's genetic symbols into pathways, which are gene groups; an acquisition unit that acquires a signature for each of the divided pathways based on a neural network learned based on the pathway that inputs each of the divided pathways; and an estimation unit that estimates whether cancer is normal or cancerous using the signature obtained for each of the pathways.

The classification unit calculates a pathway contribution index value for each of the pathways, and selects gene symbols having a preset number of high pathway contribution index values among the calculated pathway contribution index values. After that, the selected genetic symbols can be divided into the pathways.

The acquisition unit may acquire a signature for each of the pathways based on the sum of preset weights for each of the genetic symbols included in each of the pathways in the neural network.

The neural network may include pathway nodes connected one-to-one with each of the divided pathways.

The estimation unit may use the signatures obtained for each of the pathways to interpret the pathways and genetic symbols that contribute to estimating whether the cancer is normal or not.

According to embodiments of the present invention, by learning the expression patterns of genes to distinguish cancer and normal cells through deep learning, a deep neural network (DNN) structured to enable life science interpretation and understanding, the learned Using deep neural networks, prediction accuracy can be increased in cancer diagnosis, prognosis prediction, and treatment prescription, and the functions of genes related to diagnosis, prognosis, and drug effects can be numerically interpreted.

Diagnosis, prognosis, and treatment prescription based on genetic information are one of the core directions of future business, and their application is expanding, starting with the case of national health insurance in cancer. The present invention provides a deep neural network model that can learn about 20,000 pieces of publicly available gene expression information and characterize the characteristics of cancer into about 600 gene function modules, such as a gene group called Pathway. Using the underlying neural network, it is possible to determine or estimate whether something is cancerous or normal. The pathway-based learning model provided in the present invention provides the characteristics of genes that can distinguish the characteristics of cancer as about 600 numerical values that can be used by life scientists to interpret their functions, and determines cancer based on the differences between these values. It is possible to determine whether the cancer is normal and recommend or provide customized prescriptions related to the cancer.

Figure 1 shows an operation flowchart of a method for estimating cancer and normality using a neural network according to an embodiment of the present invention.
Figure 2 shows an example diagram for explaining the pathway DNN structure of the present invention.
Figure 3 shows an example comparing the cancer/normal discrimination performance of the method of the present invention and existing methods.
Figure 4 shows an example of cancer/normal discrimination accuracy according to the number of genes in the method of the present invention.
Figure 5 shows an example of cancer/normal discrimination accuracy using 57 genetic symbols selected by the present invention.
Figure 6 shows an example of AUC performance according to the number of genes in the method of the present invention.
Figure 7 shows the configuration of a cancer and normality estimation device using a neural network according to an embodiment of the present invention.

The advantages 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 accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and will be implemented in various different forms, but the present embodiments only serve to ensure that the disclosure of the present invention is complete and are within the scope of common knowledge in the technical field to which the present invention pertains. It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims.

The terminology used herein is for describing embodiments and is not intended to limit the invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context. As used herein, “comprises” and/or “comprising” refers to the presence of one or more other components, steps, operations and/or elements. or does not rule out addition.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings that can be commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not interpreted ideally or excessively unless clearly specifically defined.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. The same reference numerals are used for the same components in the drawings, and duplicate descriptions for the same components are omitted.

Embodiments of the present invention use a deep neural network (DNN) structured to enable life science interpretation and understanding by learning the expression patterns of genes to distinguish between cancer and normal cells through deep learning. The goal is to use the network to estimate or determine whether cancer is normal or not, and furthermore, to increase the accuracy of prediction for prescription of treatment for cancer.

At this time, the present invention selects a preset number of genetic symbols with a high contribution index value, for example, a pathway contribution index, to determine whether the subject's genetic symbols are cancerous or normal, and then selects the selected genetic symbols as a gene group, Pathway. By dividing it into ways and inputting each of the divided pathways into each pathway node of the neural network learned based on the pathway, it is possible to estimate or determine whether the subject has cancer or is normal using the pathway-based neural network. You can.

Furthermore, the present invention can obtain a signature for each of the pathways using a pathway-based neural network, and use the obtained signature to estimate whether it is cancerous or normal.

Figure 1 shows an operation flowchart of a method for estimating cancer and normality using a neural network according to an embodiment of the present invention.

Referring to FIG. 1, the method for estimating cancer or normality according to an embodiment of the present invention divides the subject's genetic symbols into pathways, which are gene groups (S110).

Here, a pathway refers to a gene group and can be created by combining gene symbols, for example, gene names.

At this time, step S110 calculates a pathway contribution index value for each of the pathways, and selects a preset number of the calculated pathway contribution index values, for example, 57 or 40 top pathways. After selecting genetic symbols with contribution index values, the selected genetic symbols can be divided into pathways, for example, n pathways as shown in FIG. 2.

The pathway contribution index value (P _i ) can be calculated by <Equation 1> below.

[Equation 1]

Here, g(x) means gene expression, and W(x) is the pathway node weight, which can mean the weight for each of the pathway nodes included in the neural network. there is.

For example, the most important genes in constituting the present invention may be 57, and the Pathway contribution gene index value for the 57 genes, that is, the genes distinguish between cancer and normal in the pathway layer structure. The numerical value expressing the degree of contribution may be as shown in [Table 1] below.

Gene Pathway contribution gene index PLA2G10 1.149 SHC1 0.904 CDK1 0.846 CD4 0.779 APOA1 0.721 HIST1H3A 0.663 CDC16 0.647 LHB 0.622 PLA2G2E 0.616 ORC6 0.61 HIST1H4K 0.557 GNG11 0.542 TPM1 0.537 CHRNB4 0.525 CALM1 0.52 ACAN 0.519 HLAs 0.508 HIST1H4D 0.495 PLA2G4D 0.489 MYH6 0.487 MCM6 0.479 TPM3 0.479 FGF3 0.477 CDC45 0.477 ABCG1 0.468 PPP2R1B 0.456 CTNNB1 0.453 HYAL1 0.452 PLA2G12A 0.45 NRP2 0.45 HIST1H4I 0.447 LCAT 0.443 RAC1 0.435 CD3E 0.433 MBL2 0.43 SI 0.428 GNGT2 0.427 RAP1A 0.412 CYP21A2 0.41 PRKCI 0.41 CDC25C 0.406 GNAS 0.4 STAT1 0.397 HIST2H2AA4 0.397 CHRNA3 0.396 CDH3 0.394 HAAO 0.391 OGN 0.39 PLA2G2A 0.388 PSMD13 0.383 PSMF1 0.382 PAK1 0.381 DHFR 0.379 PPP2R5D 0.376 CCNE1 0.373 CRP 0.371 HTR7 0.368

The minimum number of genes that do not deteriorate the performance of the present invention may be 40, and the 40 genes may be as shown in [Table 2] below.

Gene GNB1 MYH11 SAA1 CCR9 INS GRPR TPM2 DMD PSENEN ORAI1 SLC44A1 RELA RAC1 CXCR4 TAC1 NPY5R FGF1 EDN3 CGB8 GNGT1 TPM1 DEFB1 GLP2R PTGER3 GNAI3 SOS1 APRT MRVI1 GMPR2 DAB2 PDE3B GNG12 PRKACA CALM1 PRSS1 GSTO2 XCL1 ORC4 PDE1A SLC16A7

In step S110, when the subject's genetic symbols, for example, gene symbols selected by preprocessing the gene expression values, are divided into pathways, a neural network learned based on the pathway using each of the divided pathways as input ( For example, a signature is acquired for each of the divided pathways based on the pathway layer shown in FIG. 2 (S120).

Here, step S120 may obtain a signature for each of the pathways based on the sum of preset weights for each of the genetic symbols included in each of the pathways in the neural network.

The neural network in the present invention may be a neural network (pathway layer) learned based on the pathway shown in FIG. 2, and the output of the neural network (pathway layer) and the neural network (pathway layer) learned based on the pathway It may also be a neural network that includes fully-connected layers.

And, as shown in Figure 2, the neural network (pathway layer) learned based on the pathway may include pathway nodes connected one-to-one with each of the input pathways, and each pathway node is learned. You can have a preset weight for each of the genetic symbols.

When a signature for each of the pathways is obtained in step S120, cancer or normality is estimated using the signature obtained for each of the pathways (S130).

That is, the method according to an embodiment of the present invention uses a neural network learned based on a pathway as input, each of the pathways classified for a pre-selected number of genetic symbols, and each of the pathways in this neural network. By providing signatures for , it is possible to estimate or determine whether cancer is normal or cancerous using these signatures.

The pathway-based neural network used in the present invention may learn a neural network based on the pathway contribution index value, select the minimum gene using a gene reduction module, and then create a neural network based on the pathway using the selected minimum gene. You can also learn networks. Of course, in the present invention, the number of genetic symbols used when learning a neural network based on a pathway is not limited or restricted, and the number of genetic symbols included in the training data set is based on the performance desired by performing the method of the present invention. The number may be determined.

And, although it is shown in Figure 2 that the genetic symbols and each pathway are directly connected, it is not limited or limited thereto, and the genetic symbols are input into a separate neural network, and the genetic symbols for each of the genetic symbols in the neural network are After dividing the pathways, the classified pathways can also be input into a pathway-based neural network. That is, the present invention can input a pathway containing genetic symbols directly classified into a neural network learned based on the pathway, or can input a pathway for genetic symbols input through a neural network for distinguishing the pathway. After classification, these classified pathways can also be input into a neural network learned based on the pathway.

As such, the method according to an embodiment of the present invention uses a deep neural network (DNN) structured to enable life science interpretation and understanding by learning the expression patterns of genes to distinguish cancer and normal cells through deep learning. , using the learned deep neural network, prediction accuracy can be increased in cancer diagnosis, prognosis prediction, and treatment prescription, and the functions of genes related to diagnosis, prognosis, and drug effects can be numerically interpreted.

The present invention connects a set of genes (i.e., pathways) known to have a specific function in life science to one node in the next layer of a deep neural network, so that the meaning of the values of the genes can be summarized into one value. By learning to do so, it is possible to determine whether cancer is normal or cancerous using a neural network learned based on the pathway.

Furthermore, the method according to an embodiment of the present invention may interpret or analyze the pathways and genetic symbols that contribute to estimating cancer or normality using the signatures obtained for each of the pathways.

In other words, the present invention enables life science function analysis through the pathway index, allows knowing which pathway has explanatory power in distinguishing between cancer and normal, and can explain the distinction between cancer and normal through gene contribution.

As described above, in the present invention, the structure of a deep neural network that can implicitly interpret expression patterns consisting of about 5,700 genes as a unit of a certain number of gene functions considered important in life science, such as about 600, and By providing the values of the weights that make up the structure and securing high performance using a deep neural network, we can provide cancer/normal diagnosis, prognosis prediction, and drug efficacy prediction methods based on multivariate discriminants using genes. You can.

In addition, in that the expression patterns of about 5,700 genes by the structural pattern of the present invention can be numerically explained with about 600 functional modules known in life science, it provides life science explanations that ordinary deep neural networks cannot provide. can be presented.

Figure 3 shows an example diagram comparing the cancer/normal discrimination performance of the method of the present invention and existing methods, showing the deep neural network discrimination performance of cancer/normal using gene expression pattern information.

As can be seen from Figure 3, when the method of the present invention is used, a fully connected deep neural network using the expression value of the same gene as that of the present invention, and a deep neural network using the expression value of the same gene as that of the present invention and a random layer. It can be seen that excellent performance can be secured.

Figure 4 shows an example of the accuracy of cancer/normal discrimination according to the number of genes in the method of the present invention. As shown in Figure 4, the accuracy of cancer/normal discrimination when all 20,000 genes are used (100%). is 99.31%, and when 2350 genes are used (10%), the accuracy of cancer/normal discrimination is 99.25%, and when 57 genes are used (2%), the accuracy of cancer/normal discrimination is 97.21%, 57 It is possible to accurately distinguish between cancer and normal dogs using just their genes. In addition, it can be seen that the method of the present invention does not significantly reduce the performance of distinguishing between cancer and normal genes in the top 1% of genes with high pathway contribution index values.

Figure 5 shows an example of cancer/normal discrimination accuracy using 57 genetic symbols selected by the present invention.

As can be seen from Figure 5, it can be seen that cancer/normal discrimination using the method of the present invention using the 57 major genes listed in Table 1 provides significantly better performance compared to the method using the same number of randomly selected genes. there is. In other words, it can be seen that the ability to distinguish between cancer and normal of 57 genes with high pathway contribution index values is better than that of 57 genes selected at random.

Figure 6 shows an example of AUC performance according to the number of genes in the method of the present invention. Based on the AUC performance, genes that do not affect performance are excluded and a neural network learned with the remaining genes is created. This was repeated until performance was maintained.

As can be seen in Figure 6, the performance for all genes is 0.9994 based on AUC, and when there are 40 genes remaining, the performance is 0.9964 based on AUC, showing that the AUC performance is maintained.

Figure 7 shows the configuration of a device for estimating cancer and normality using a neural network according to an embodiment of the present invention, and shows the conceptual configuration of a device that performs the method of Figures 1 to 6.

Referring to FIG. 7, the cancer/normality estimation device 700 according to an embodiment of the present invention includes a classification unit 710, an acquisition unit 720, and an estimation unit 730.

The classification unit 710 divides the subject's genetic symbols into pathways, which are gene groups.

At this time, the classification unit 710 calculates a pathway contribution index value for each of the pathways, and selects a preset number of the calculated pathway contribution index values, for example, 57 or 40. After selecting gene symbols with high pathway contribution index values, the selected gene symbols can be divided into pathways.

The acquisition unit 720 acquires a signature for each of the divided pathways based on a neural network learned based on a pathway that uses each of the pathways classified by the dividing unit 710 as input.

Here, the acquisition unit 720 may acquire a signature for each of the pathways based on the sum of preset weights for each of the genetic symbols included in each of the pathways in the neural network.

The estimation unit 730 uses the signature obtained for each of the pathways to estimate whether the pathway is cancerous or normal.

Furthermore, the estimation unit 730 can interpret the pathways and genetic symbols that contribute to estimating cancer or normality using the signatures obtained for each of the pathways.

Although the description is omitted in the device of FIG. 7, it is obvious to those skilled in the art that the device according to the present invention can include all the contents described in FIGS. 1 to 6 above.

The system or device described above may be implemented with hardware components, software components, and/or a combination of hardware components and software components. For example, the systems, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), etc. ), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. A processing device may execute an operating system (OS) and one or more software applications that run 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 multiple 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.

Methods according to embodiments 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 of the claims also fall within the scope of the claims described below.

Claims (11)

Classifying the subject's genetic symbols into pathways, which are gene groups;
Obtaining a signature for each of the divided pathways based on a neural network learned based on the pathway using each of the divided pathways as input; and
Step of estimating whether cancer is normal or not using the signature obtained for each of the pathways
Including,
The steps for dividing into the above pathways are
PLA2G10, SHC1, CDK1, CD4, APOA1, HIST1H3A, CDC16, LHB, PLA2G2E, ORC6, HIST1H4K, GNG11, TPM1, CHRNB4, CALM1, ACAN, HLA, HIST1H4D, PLA2G4D, MYH6, MCM6, TPM3, FGF3, CDC45, ABCG1, PPP2R1B, CTNNB1, HYAL1, PLA2G12A, NRP2, HIST1H4I, LCAT, RAC1, CD3E, MBL2, SI, GNGT2, RAP1A, CYP21A2, PRKCI, CDC25C, GNAS, STAT1, HIST2H2AA4, CHRNA3, CDH3, HAAO, OGN, PLA2G2A, PSMD13, Among the 57 gene symbols including PSMF1, PAK1, DHFR, PPP2R5D, CCNE1, CRP, and HTR7, a pathway contributing gene index indicating the degree of contribution to the distinction between cancer and normal was utilized for each of at least 40 gene symbols. Method for estimating cancer and normal status using a characteristic neural network. According to paragraph 1,
The steps for dividing into the above pathways are
For each of the pathways, a pathway contribution gene index value is calculated, and gene symbols having a preset number of high pathway contribution gene index values among the calculated pathway contribution gene index values are selected. A method for estimating cancer and normality using a neural network, characterized in that after selection, the selected genetic symbols are divided into the pathways.
According to paragraph 1,
The step of acquiring the signature is
Estimation of cancer and normality using a neural network, characterized in that a signature for each of the pathways is obtained based on the sum of preset weights for each of the genetic symbols included in each of the pathways in the neural network. method.
According to paragraph 1,
The neural network is
A method for estimating cancer and normalcy using a neural network, comprising pathway nodes connected one-to-one with each of the divided pathways.
According to paragraph 1,
A step of interpreting the pathways and genetic symbols that contribute to estimating whether the cancer is normal or not using the signatures obtained for each of the pathways.
A method for estimating cancer and normal status using a neural network further comprising:
delete delete delete delete delete delete

Images