Anticancer drug response prediction method and system based on multi-view graph comparison learning framework

Abstract

The invention belongs to but is not limited to the technical field of multi-omics data prediction, and particularly relates to an anti-cancer drug response prediction method and system based on a multi-view graph comparison learning framework. An omics-graph conversion module converts original multi-omics data into multi-view graph data through an omics-graph conversion algorithm; the de-noising module processes noise in the multi-view image data through a de-noising algorithm; the graph comparison learning framework utilizes gene interaction information in the STRING database to compare with the processed multi-view graph data; the loss function adopts Hausdorff similarity measurement to optimize the expression of the multi-view graph data; the drug data and the multi-view graph data of the cells are input into an output module together, and a prediction result is output through GNN and a full-connection network. The method can effectively predict the combination condition of the anti-cancer drug and the cancer cells, and the prediction accuracy is improved.

Description

Technical Field

[0001] The present invention belongs to but is not limited to the field of multi-omics data prediction technology, and in particular relates to an anticancer drug response prediction method and system based on a multi-view graph comparison learning framework. Background Art

[0002] Cancer has become a major global public health challenge and a major obstacle to improving life expectancy in the 21st century. Identifying effective anticancer drugs is crucial to improving patients' chances of successful recovery. Recently, various computational models have been developed and used to predict drug responses in hundreds of cell lines with encouraging results. With the continuous development of biotechnology, genomic, transcriptomic and proteomic analysis has become the focus of cancer research, and these omics data are essential for characterizing cancer phenotypes and gene features with therapeutic impact. However, most existing prediction methods find it difficult to effectively integrate multi-omics data from cancer cell lines for prediction.

[0003] Current methods have the following defects: (1) Current prediction methods are highly dependent on manual feature screening, but the feature dimensions of multi-omics data are high, and manual screening of appropriate features is not feasible; (2) In view of the noise existing in multi-omics data, traditional prediction methods lack data filtering modules. (3) Traditional prediction methods lack feature mining between multi-omics data. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides an anticancer drug response prediction method and system based on a multi-view graph comparison learning framework.

[0005] The present invention is implemented as follows: a method for predicting anticancer drug response based on a multi-view graph comparison learning framework, comprising the following steps:

[0006] Step 1, the omics-to-graph conversion module converts the original multi-omics data into multi-view graph data through the omics-to-graph conversion algorithm;

[0007] Step 2: The denoising module processes the noise in the multi-view image data through a denoising algorithm;

[0008] Step 3: The graph comparison learning framework uses the gene interaction information in the STRING database to compare with the processed multi-view graph data; the loss function uses the Hausdorff similarity metric to optimize the representation of the multi-view graph data;

[0009] Step 4: The drug data and the multi-view image data of the cell are input into the output module together, and the prediction results are output through the GNN and the fully connected network.

[0010] Furthermore, the multi-omics data input Input consists of V1,…V nThe expression matrix of different omics is composed of n, where n represents the number of omics used, and the expression matrix V n The shape is n_Cell*n_Gene, n_Cell represents the number of cells, and n_Gene represents the number of genes.

[0011] Furthermore, the omics-graph conversion algorithm is:

[0012] (1) For any input omics V n The Omics-Graph Conversion module calculates the Pearson correlation coefficient matrix P of the expression of n_Gene genes in n_Cell cells. n ;

[0013] (2) Setting the threshold αn in advance; correlation coefficient matrix P n Less than α n The threshold value is set to 0, and the value greater than or equal to the threshold value is set to 1 to obtain the omics V n The adjacency matrix of

[0014] (3) Repeat steps (1)-(2) until the adjacency matrix is ​​obtained for each type of omics data to form multi-view graph structure data.

[0015] Furthermore, the denoising algorithm is the T-SVD method.

[0016] Furthermore, the graph comparison learning framework includes a multi-view graph feature extraction module, a graph data comparison module, and a prediction input module.

[0017] Furthermore, the multi-view graph feature extraction module uses adaptive residual connection to extract multi-view graph features, and the graph data comparison module uses Hausdorff distance to optimize multi-view graph features; the prediction input module predicts the results based on the multi-view graph features and drug molecular features.

[0018] Another object of the present invention is to provide an anticancer drug response prediction method based on a multi-view graph comparison learning framework and an anticancer drug response prediction system based on a multi-view graph comparison learning framework, comprising:

[0019] The multi-omics-to-graph data conversion module converts the original multi-omics data into multi-view graph data through the omics-to-graph conversion algorithm;

[0020] Multi-omics data denoising module, which processes the noise in multi-view graph data through denoising algorithm;

[0021] The graph comparison learning framework uses the gene interaction information in the STRING database to compare with the processed multi-view graph data; the loss function uses the Hausdorff similarity metric to optimize the representation of multi-view graph data;

[0022] Output module, drug data and multi-view image data of cells are input into the output module together, and the prediction results are output through GNN and fully connected network.

[0023] Another object of the present invention is to provide a computer device, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the anticancer drug response prediction method based on a multi-view graph comparison learning framework.

[0024] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the anticancer drug response prediction method based on a multi-view graph comparison learning framework.

[0025] Another object of the present invention is to provide an information data processing terminal, which includes the anticancer drug reaction prediction system based on the multi-view graph comparison learning framework.

[0026] In combination with the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solutions to be protected by the present invention are as follows:

[0027] First, the multi-omics data mapping method provided by the present invention can convert multi-omics data into multi-view graph structure data, which is convenient for the use of graph structure-based prediction methods and improves prediction performance; the denoising module can remove the noise in the sequencing data and improve the data quality.

[0028] The graph comparison learning framework provided by the present invention introduces gene interaction information existing in the database to assist prediction, which can effectively predict the binding of anticancer drugs to cancer cells and improve the prediction accuracy.

[0029] Second, the technical solution of the present invention fills the technical gap in the industry at home and abroad: the current processing method for multi-omics data usually studies each omics data separately and directly integrates the processing results; while the technical solution of the present invention can directly integrate multi-omics data and convert multi-omics data into graphs, which can effectively mine the internal associations of multi-omics data. In addition, the T-SVD technical solution of the present invention is also the first application of denoising technology for omics data.

[0030] Third, the technical solution of the present invention solves the problem of multi-omics data analysis in the prior art in industrial applications, especially in the prediction of anticancer drug response. By introducing a multi-view graph comparison learning framework, it overcomes the problems of large data noise, insufficient feature extraction, and low prediction accuracy in traditional methods. Specifically, by converting multi-omics data into multi-view graph form and combining it with a denoising algorithm, noise interference in the data is eliminated, significantly improving the accuracy and robustness of data processing.

[0031] In addition, the present invention optimizes multi-view graph features using Hausdorff distance and combines graph neural network (GNN) and fully connected network (FCN) for joint prediction, which not only improves the prediction accuracy of anticancer drug response but also provides interpretable prediction results. This method effectively solves the problem that traditional models cannot fully capture complex gene interactions when processing multi-omics data.

[0032] Significant technological progress is also reflected in the efficiency and applicability of this method. Compared with the existing single-omics data processing method, this invention achieves a global evaluation of cell and drug response through multi-omics fusion and feature optimization, which is suitable for a wider range of anti-cancer drug development and clinical diagnosis scenarios, and promotes the further application of precision medicine in the field of anti-cancer. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 is a flow chart of an anticancer drug response prediction method based on a multi-view graph comparison learning framework provided by an embodiment of the present invention;

[0034] Figure 2 It is a structural diagram of an anticancer drug reaction prediction system based on a multi-view graph comparison learning framework provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0036] like Figure 1 As shown, an anticancer drug response prediction method based on a multi-view graph comparison learning framework provided by an embodiment of the present invention comprises the following steps:

[0037] Step 1, the omics-to-graph conversion module converts the original multi-omics data into multi-view graph data through the omics-to-graph conversion algorithm;

[0038] Step 2: The denoising module processes the noise in the multi-view image data through a denoising algorithm;

[0039] Step 3: The graph comparison learning framework uses the gene interaction information in the STRING database to compare with the processed multi-view graph data; the loss function uses the Hausdorff similarity metric to optimize the representation of the multi-view graph data;

[0040] Step 4: The drug data and the multi-view image data of the cell are input into the output module together, and the prediction results are output through the GNN and the fully connected network.

[0041] The embodiment of the present invention receives a multi-omics input Input, consisting of V1, ...V n The expression matrix of different omics is composed of n, where n represents the number of omics used, and the expression matrix V n The shape is n_Cell*n_Gene, n_Cell represents the number of cells, and n_Gene represents the number of genes.

[0042] The omics-graph conversion algorithm is:

[0043] (1) For any input omics V n The Omics-Graph Conversion module calculates the Pearson correlation coefficient matrix P of the expression of n_Gene genes in n_Cell cells. n ;

[0044] (2) Setting the threshold αn in advance; correlation coefficient matrix P n Less than α n The threshold value is set to 0, and the value greater than or equal to the threshold value is set to 1 to obtain the omics V n The adjacency matrix of

[0045] (3) Repeat steps (1)-(2) until the adjacency matrix is ​​obtained for each type of omics data to form multi-view graph structure data.

[0046] The embodiment of the present invention provides a multi-omics data mapping method, including a multi-omics-graph data conversion module and a multi-omics data denoising module. The original multi-omics data is converted into a multi-view graph by the multi-omics-graph data conversion module. The multi-omics data denoising module processes the noise in the multi-view graph to improve the quality of the input data. The denoising algorithm uses the T-SVD method to eliminate the noise in the multi-view graph data.

[0047] The embodiment of the present invention provides a graph comparison learning framework, which predicts the binding status of anticancer drugs to different cells based on omics multi-view graphs and molecular graphs.

[0048] The graph comparison learning framework includes a multi-view graph feature extraction module, a graph data comparison module, and a prediction input module. The multi-view data uses adaptive residual connections to extract multi-view graph features. The graph data comparison module uses Hausdorff distance to optimize multi-view graph features. The prediction input module predicts the results based on the multi-view graph features and drug molecular features.

[0049] like Figure 2 As shown, an anticancer drug reaction prediction method based on a multi-view graph comparison learning framework and an anticancer drug reaction prediction system based on a multi-view graph comparison learning framework provided by an embodiment of the present invention include:

[0050] The multi-omics-to-graph data conversion module converts the original multi-omics data into multi-view graph data through the omics-to-graph conversion algorithm;

[0051] Multi-omics data denoising module, which processes the noise in multi-view graph data through denoising algorithm;

[0052] The graph comparison learning framework uses the gene interaction information in the STRING database to compare with the processed multi-view graph data; the loss function uses the Hausdorff similarity metric to optimize the representation of multi-view graph data;

[0053] Output module, drug data and multi-view image data of cells are input into the output module together, and the prediction results are output through GNN and fully connected network.

[0054] The anticancer drug response prediction system based on the multi-view graph contrast learning framework provided by the embodiment of the present invention aims to improve the accuracy and reliability of anticancer drug response prediction by combining multi-omics data with graph contrast learning technology. The system uses omics data, graph contrast learning framework and gene interaction database to achieve accurate prediction of drug response.

[0055] Step 1: Multi-omics-graph data conversion

[0056] First, the original multi-omics data is converted into multi-view graph data through the multi-omics-graph data conversion module. Multi-omics data includes information such as gene expression, DNA methylation, and proteomics. These multi-dimensional data are processed by the omics-graph conversion algorithm to generate corresponding graph data. Each view graph data represents different biological level information. In this way, a multi-view graph structure that reflects the internal mechanism of the cell and the interaction of genes can be constructed, providing a data basis for the subsequent learning framework.

[0057] Step 2: Multi-omics data denoising

[0058] Since multi-omics data usually contains noise and redundant information, which affects the accuracy of the prediction results, the present invention uses a multi-omics data denoising module for preprocessing. This module uses a denoising algorithm to process the noise of the generated multi-view graph data, removes interference information, and retains key biometric data. This step ensures that the subsequent comparative learning framework can run on a cleaner and more accurate basis, thereby improving the prediction effect of the model.

[0059] Step 3: Construction of graph contrast learning framework

[0060] The core part of the system is the graph contrast learning framework. This framework combines the gene interaction information from the STRING database and compares the processed multi-view graph data with the gene interaction network. Through this contrast learning method, the system is able to extract important features in the prediction of anticancer drug response and optimize the expression of multi-view graph data based on gene network information. This contrast mechanism can help the model better capture the complex relationship between drugs and cells.

[0061] Step 4: Loss Function and Optimization

[0062] In the graph contrast learning framework, the loss function uses the Hausdorff similarity metric to measure the similarity differences between different views. Hausdorff distance is a metric commonly used to calculate the similarity between two graphs. By minimizing the Hausdorff distance between the view graph data and the gene interaction graph, the expression of multi-view graph data can be optimized and the model's adaptability to complex biological data can be enhanced. This optimization process helps the system learn more effective feature representations and improve the accuracy of drug response prediction.

[0063] Step 5: Prediction model construction and reasoning

[0064] Based on the graph contrast learning framework, drug data and processed cell multi-view graph data are input into the prediction module. The prediction module performs in-depth feature extraction and association analysis on these multi-view graph data through the combination of graph neural network (GNN) and fully connected neural network (FCN). GNN can process the relationship information in graph structure data, while FCN is used to further extract high-level features and implement classification or regression tasks. Finally, the prediction module outputs the prediction results of drug response to cancer cells.

[0065] Step 6: System output and application

[0066] The final output module of the system is responsible for converting the results of the multi-view data analysis of drugs and cells into prediction results for reference by researchers or clinicians. Through this system, users can get predictions of the response of anticancer drugs to specific cancer cells, assist in drug selection and formulate personalized treatment plans. This method not only improves the accuracy of predictions, but also better adapts to different types of omics data and is suitable for a wide range of applications in the field of cancer treatment.

[0067] This paper proposes an anticancer drug response prediction method based on a multi-view graph contrast learning framework. By combining multi-omics data with graph neural network (GNN) technology, it realizes efficient prediction of anticancer drug responses to different cells. First, the multi-omics data is converted into a multi-view graph form through the omics-graph conversion module to provide structured data representation for subsequent graph contrast learning and prediction. This framework mainly includes four steps: omics-graph conversion, denoising, graph contrast learning and final output prediction.

[0068] In step one, the original multi-omics data (such as gene expression, DNA methylation, etc.) are input into the omics-graph conversion module. Each omics data is processed into an expression matrix, where rows represent cells and columns represent genes. Then, the omics-graph conversion algorithm generates a correlation coefficient matrix by calculating the Pearson correlation coefficient between genes. By setting a threshold, the elements with correlation below the threshold are set to 0, and the elements with correlation above the threshold are set to 1, and the adjacency matrix of each omics data is constructed to form a multi-view graph structure data.

[0069] In step 2, in order to improve the accuracy of the multi-view graph data, the denoising module uses the T-SVD (truncated singular value decomposition) method to process the noise in the data. This algorithm performs matrix decomposition on the original data, retains the main components and removes noise interference. This step ensures that the data used in the subsequent prediction step is more accurate and reduces the impact of noise interference on the prediction model.

[0070] In step three, a graph comparison learning framework is introduced to optimize the representation of multi-view graph data. The framework uses the gene interaction information in the STRING database as a benchmark graph to compare with the multi-view graph data. The features of different omics graphs are extracted through the multi-view graph feature extraction module. In order to optimize the expression accuracy, the Hausdorff distance is used as the similarity metric function to minimize the difference between the different view graph data and the benchmark graph. The graph data comparison module optimizes feature extraction based on this similarity, ensuring that the multi-view graph performs more reliably in the prediction of anticancer drug response.

[0071] In step 4, the multi-view graph data after denoising and contrast learning is input into the prediction module, and the drug data is input together with the processed cell multi-view graph data. Through the combination of graph neural network (GNN) and fully connected network (FCN), the model can extract rich cell features from the graph data, and combine the drug molecular features to predict the anticancer drug response. GNN plays the role of extracting multi-dimensional features in this process, and FCN is responsible for the final prediction result output.

[0072] Through the joint input of multi-view graph features and drug molecular features, the system can output the prediction results of drugs on cells. The prediction results not only contain information about the drug's response to specific cells, but also can evaluate the sensitivity or drug resistance of cells, providing data support for personalized treatment. In addition, through the synergy of GNN and fully connected networks, this framework not only improves the accuracy of prediction, but also generates interpretable graph structures, providing technical references for subsequent drug development and clinical applications.

[0073] This method enhances the expression ability of multi-omics data by introducing multi-view graph contrast learning and denoising, optimizes feature extraction using Hausdorff distance, and combines GNN and FCN to predict anticancer drug response. Through the joint analysis of multi-view graph features and drug molecular features, this method provides an efficient and reliable prediction method for drug development and clinical diagnosis, and promotes the development of precision medicine.

[0074] The present invention is compared with the current mainstream anticancer drug cell binding prediction method on the GDSC2 dataset, and three indicators, RSME (Root Mean Squared Error), MAE (Mean Absolute Error) and PCC (Pearson correlation coefficient), are evaluated. The results are shown in the following figure.

[0075]

[0076] It can be seen that the method (TGCCL) of this scheme has achieved the best prediction performance in three different indicators compared with other methods. This fully demonstrates the effectiveness of this scheme. At the same time, we conducted an ablation experiment, removing the denoising module T-SVD, the adaptive residual module Res, and the image comparison module GCL in this scheme, and compared them with the original results of this scheme. The results are shown in the following figure.

[0077]

[0078] Through ablation experiments, it is fully proved that the different modules proposed in this scheme can improve the prediction performance, and further verify the effects of each part of this scheme.

[0079] It should be noted that the embodiments of the present invention can be implemented by hardware, software, or a combination of software and hardware. The hardware part can be implemented using dedicated logic; the software part can be stored in a memory and executed by an appropriate instruction execution system, such as a microprocessor or dedicated design hardware. It can be understood by a person of ordinary skill in the art that the above-mentioned devices and methods can be implemented using computer executable instructions and / or contained in a processor control code, such as a carrier medium such as a disk, CD or DVD-ROM, a programmable memory such as a read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. Such code is provided on the carrier medium. The device and its modules of the present invention can be implemented by hardware circuits such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, etc., or programmable hardware devices such as field programmable gate arrays, programmable logic devices, etc., can also be implemented by software executed by various types of processors, and can also be implemented by a combination of the above-mentioned hardware circuits and software, such as firmware.

[0080] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any modifications, equivalent substitutions and improvements made by any technician familiar with the technical field within the technical scope disclosed by the present invention and within the spirit and principle of the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for predicting anticancer drug response based on a multi-view graph contrast learning framework, characterized in that: The following steps are involved: Step 1, the omics-to-graph conversion module converts the original multi-omics data into multi-view graph data through the omics-to-graph conversion algorithm; Step 2: The denoising module processes the noise in the multi-view image data through a denoising algorithm; Step 3: The graph comparison learning framework uses the gene interaction information in the STRING database to compare with the processed multi-view graph data; The loss function uses Hausdorff similarity metric to optimize the representation of multi-view graph data; Step 4: The drug data and the multi-view image data of the cell are input into the output module together, and the prediction results are output through the GNN and the fully connected network.

2. The method for predicting anticancer drug response based on a multi-view graph contrast learning framework according to claim 1, characterized in that: Multi-omics data input Input consists of V1,…V n The expression matrix of different omics is composed of n, where n represents the number of omics used, and the expression matrix V n The shape is n_Cell*n_Gene, n_Cell represents the number of cells, and n_Gene represents the number of genes.

3. The method for predicting anticancer drug response based on a multi-view graph contrast learning framework according to claim 1, characterized in that: The omics-graph conversion algorithm is: (1) For any input omics V n The Omics-Graph Conversion module calculates the Pearson correlation coefficient matrix P of the expression of n_Gene genes in n_Cell cells. n ; (2) Setting the threshold αn in advance; correlation coefficient matrix P n Less than α n The threshold value is set to 0, and the value greater than or equal to the threshold value is set to 1 to obtain the omics V n The adjacency matrix of (3) Repeat steps (1)-(2) until the adjacency matrix is ​​obtained for each type of omics data to form multi-view graph structure data.

4. The method for predicting anticancer drug response based on a multi-view graph contrast learning framework according to claim 1, characterized in that: The denoising algorithm is the T-SVD method.

5. The method for predicting anticancer drug response based on a multi-view graph contrast learning framework according to claim 1, characterized in that: The graph comparison learning framework includes a multi-view graph feature extraction module, a graph data comparison module, and a prediction input module.

6. The method for predicting anticancer drug response based on a multi-view graph contrast learning framework according to claim 1, characterized in that: The multi-view graph feature extraction module uses adaptive residual connection to extract multi-view graph features, and the graph data comparison module uses Hausdorff distance to optimize multi-view graph features. The prediction input module predicts the results based on the multi-view graph features and drug molecular features.

7. An anticancer drug response prediction system based on a multi-view graph comparison learning framework according to any one of claims 1 to 6, characterized in that: include: The multi-omics-to-graph data conversion module converts the original multi-omics data into multi-view graph data through the omics-to-graph conversion algorithm; Multi-omics data denoising module, which processes the noise in multi-view graph data through denoising algorithm; A graph comparison learning framework that uses gene interaction information from the STRING database to compare with processed multi-view graph data; The loss function uses Hausdorff similarity metric to optimize the representation of multi-view graph data; Output module, drug data and multi-view image data of cells are input into the output module together, and the prediction results are output through GNN and fully connected network.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the anticancer drug response prediction method based on a multi-view graph comparison learning framework as described in any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to execute the steps of the anticancer drug response prediction method based on a multi-view graph contrast learning framework as described in any one of claims 1 to 6.

10. An information data processing terminal, characterized in that: The information data processing terminal includes the anticancer drug reaction prediction system based on the multi-view graph comparison learning framework described in claim 7.

Images