Drug sensitivity classification method, system and equipment based on multi-omics data integration

Abstract

The present invention discloses a drug sensitivity classification method, system and equipment based on multi-omics data integration, which relates to the field of drug sensitivity classification. The method of the present invention uses joint non-negative matrix decomposition to decompose multiple omics matrices of all multi-omics data samples, and the obtained shared matrix represents the linear integration features of all samples; at the same time, a variational autoencoder (VAE) is used to obtain the nonlinear integration features of all samples; the spliced ​​feature vector obtained by splicing the linear integration features, the nonlinear integration features and the drug features is used for drug sensitivity classification. The linear integration features contain important information of all omics and eliminate the noise in the original data. The nonlinear integration features contain nonlinear and complex relationship information in the multi-omics data. The present invention utilizes the complementarity of the two features to mine the potential information between the multi-omics data, thereby greatly improving the accuracy of drug sensitivity classification.

Description

Technical Field

[0001] The present invention relates to the technical field of drug sensitivity classification, and in particular to a drug sensitivity classification method, system and device based on multi-omics data integration. Background Art

[0002] Drug sensitivity classification is a crucial step in drug design and discovery. Because drug design and discovery is an extremely lengthy process, drug sensitivity classification is both a significant and challenging task. Accurate drug sensitivity classification facilitates the selection of promising drugs and aids drug screening.

[0003] With the development of medical radiomics and high-throughput technology, multi-omics data has exploded in modern biomedical research. For the same biological sample, different biotechniques can be used to obtain different omics information. Omic data include gene expression, DNA methylation, etc. Although single-omics data can describe biological characteristics from a certain perspective, the information it contains is very limited and cannot describe the subtle differences of organisms. Integrating multi-omics data can more comprehensively describe biological characteristics from multiple perspectives at the same time. Existing studies have shown that multi-omics data plays an important role in drug sensitivity classification. Compared with using only single-omics data, integrating multi-omics data can significantly improve the classification accuracy of drug sensitivity. Therefore, it is very necessary to find a method that can effectively integrate multi-omics data and classify drug sensitivity.

[0004] In the field of multi-omics data integration, existing methods are mostly based on traditional machine learning and deep learning methods. Traditional machine learning methods, such as matrix decomposition, focus solely on linear relationships between multi-omics features while ignoring nonlinear relationships between features. While deep learning-based methods can better capture nonlinear information and complex relationships in multi-omics data, existing methods often fail to utilize both linear and nonlinear integration features of multi-omics data, resulting in an inability to fully exploit the potential information of multi-omics data, which in turn affects classification performance. Summary of the Invention

[0005] In response to the problems raised in the above background technology, the present invention provides a drug sensitivity classification method, system and device based on multi-omics data integration to greatly improve the accuracy of drug sensitivity classification.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] In one aspect, the present invention provides a drug sensitivity classification method based on multi-omics data integration, comprising:

[0008] Obtain multi-omics data samples by sequencing cell lines and construct multiple omics matrices;

[0009] Decomposing the multiple omics matrices using joint non-negative matrix factorization, and adding graph regularization to the matrix decomposition by constructing structural similarity of the multi-omics data samples to obtain linear integration features of the multi-omics data samples;

[0010] Use variational autoencoders to obtain nonlinear integration features of multi-omics data samples;

[0011] splicing the linear integration features, the nonlinear integration features, and the drug features of the multi-omics data sample to obtain a spliced ​​feature vector;

[0012] training a drug sensitivity classifier using the concatenated feature vector;

[0013] The trained drug sensitivity classifier is used to perform drug sensitivity classification to obtain the classification results.

[0014] Optionally, obtaining multi-omics data samples by sequencing cell lines and forming multiple omics matrices specifically includes:

[0015] Obtaining multi-omics data samples by sequencing cell lines M is the number of multi-omics data samples obtained; To form the multi-omics data sample x i The k-th omics data, 1≤k≤K; K is the multi-omics data sample x i The number of omics data included in the ; the omics data include gene expression and DNA methylation;

[0016] The M multi-omics data samples obtained are used to form K omics matrices X = {X (1) ,…,X (K)}; where the k-th omics matrix

[0017] Optionally, the multiple omics matrices are decomposed using joint non-negative matrix factorization, and graph regularization is added to the matrix decomposition by constructing structural similarity of the multi-omics data samples to obtain linear integration features of the multi-omics data samples, specifically including:

[0018] The K omics matrices X = {X (1) ,…,X (K)}Decomposed into a shared matrix U and K transformation matrices V (1) ,…,V (K) In the decomposition process, the structural similarity of multi-omics data samples is constructed to add graph regularization to the matrix decomposition, and the row vector u of the shared matrix U obtained by decomposition is i As a multi-omics data sample xi Linear integration characteristics.

[0019] Optionally, the method of obtaining nonlinear integration features of multi-omics data samples using a variational autoencoder specifically includes:

[0020] Using the variational autoencoder encoder f encoder , using the formula Generate multi-omics data sample x i The corresponding mean vector μ i and the standard deviation vector σ i ;

[0021] The obtained mean vector μ i As a multi-omics data sample x i Nonlinear integration characteristics.

[0022] Optionally, the step of splicing the linear integration features, the nonlinear integration features, and the drug features of the multi-omics data sample to obtain a spliced ​​feature vector specifically includes:

[0023] The multi-omics data sample x i The linear integration characteristic u i , nonlinear integration characteristics μ i The drug feature d is directly concatenated head to tail to obtain the concatenated feature vector.

[0024] Optionally, the method of training a drug sensitivity classifier using the concatenated feature vector specifically includes:

[0025] The concatenated feature vector is used to train a drug sensitivity classifier, and the training process adopts a 5-fold cross-validation method for training and testing to obtain a trained drug sensitivity classifier; the input of the drug sensitivity classifier is the concatenated feature vector, and the output is the drug sensitivity category, which includes sensitive, resistant and intermediate areas.

[0026] In another aspect, the present invention provides a drug sensitivity classification system based on multi-omics data integration, comprising:

[0027] A sample collection module is used to obtain multi-omics data samples by sequencing cell lines and construct multiple omics matrices;

[0028] a linear integration module for decomposing the multiple omics matrices using joint non-negative matrix factorization, and adding graph regularization to the matrix decomposition by constructing structural similarity of the multi-omics data samples to obtain linear integration features of the multi-omics data samples;

[0029] Nonlinear integration module, used to obtain nonlinear integration features of multi-omics data samples using variational autoencoders;

[0030] A feature splicing module, configured to splice the linear integration features, nonlinear integration features, and drug features of the multi-omics data sample to obtain a spliced ​​feature vector;

[0031] a classifier training module, configured to train a drug sensitivity classifier using the concatenated feature vector;

[0032] The sensitivity classification module is used to perform drug sensitivity classification using a trained drug sensitivity classifier to obtain classification results.

[0033] On the other hand, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the drug sensitivity classification method based on multi-omics data integration when executing the computer program.

[0034] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which implements the drug sensitivity classification method based on multi-omics data integration when the computer program is executed.

[0035] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0036] The present invention provides a drug sensitivity classification method, system, and device based on multi-omics data integration. The method uses joint non-negative matrix decomposition to decompose multiple omics matrices of all multi-omics data samples. The resulting shared matrix represents the linear integration features of all samples. A variational autoencoder (VAE) is used to obtain nonlinear integration features of all samples. The concatenated feature vector obtained by concatenating the linear integration features, nonlinear integration features, and drug features is used for drug sensitivity classification. The linear integration features contain important information from all omics, eliminating noise in the original data. The nonlinear integration features contain nonlinear and complex relationship information in the multi-omics data. The present invention utilizes the complementarity of the two features to mine potential information between multi-omics data, thereby greatly improving the accuracy of drug sensitivity classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1This is a flow chart of a drug sensitivity classification method based on multi-omics data integration provided by the present invention. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] The purpose of the present invention is to provide a drug sensitivity classification method, system and equipment based on multi-omics data integration to greatly improve the accuracy of drug sensitivity classification.

[0041] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] Figure 1 This is a flow chart of a drug sensitivity classification method based on multi-omics data integration provided by the present invention, see Figure 1 , a drug sensitivity classification method based on the integration of multi-omics data, including:

[0043] Step 1: Obtain multi-omics data samples by sequencing cell lines and construct multiple omics matrices.

[0044] The multi-omics data sample used in the present invention is derived from the sequencing of biological sample cell lines. A multi-omics data sample (also referred to as a multi-omics sample) x i (1≤i≤M) consists of K omics data, that is, Composition of multi-omics data sample x i The k-th omics data in, represents real numbers, Indicated by N (k) A one-dimensional vector of real numbers. It constitutes the multi-omics data sample x i Omics data include but are not limited to gene expression, DNA methylation, etc., which are usually obtained using specialized sequencing technologies.

[0045] For a dataset with M multi-omics samples, all multi-omics samples can be represented as K omics matrices X = {X (1) ,…,X (K)}, where the k-th omics matrix Each multi-omics sample x i It can form the final sample pair with the drug sample d, the sample pair {x i,d} is encoded with a one-hot encoding vector y i ∈{0,1} 1×C Indicates, C represents the number of all sample pairs. Drug sample d (e.g. poly ADP-ribose polymerase (PARP)) is represented by the molecular structure of the drug, which is also used as its drug feature and is generally obtained from the drug structure database. Sample pair {x i ,d}’s true types include “sensitive”, “resistant” and “intermediate region”.

[0046] Step 2: Decompose the multiple omics matrices using joint non-negative matrix factorization, and add graph regularization to the matrix decomposition by constructing the structural similarity of the multi-omics data samples to obtain the linear integration features of the multi-omics data samples.

[0047] The present invention uses joint non-negative matrix decomposition to decompose the K omics matrices of all multi-omics samples, and the resulting shared matrix represents the linear integration characteristics of all multi-omics samples. The decomposition process is to convert the K omics matrices {X (1) ,…,X (K)} decomposed into a shared matrix and multiple transformation matrices V (1) ,…,V (K) , where the kth transformation matrix The row vector u of U i (1≤i≤M) as each sample x i The linear integration characteristics of , the decomposition expression is as follows:

[0048]

[0049] Where D represents the dimension of the linear integration feature after decomposition, represents a non-negative real matrix of dimension M×D, Represents a dimension of D×N (k) Non-negative real matrix, N (k) represents the original dimension of the k-th omics data, stands for the Frobenius norm squared.

[0050] In order to learn sample similarity information in omics data, the present invention adds graph regularization to matrix decomposition by constructing the structural similarity (SSIM) of multi-omics samples. The specific process is as follows:

[0051] First, calculate the similarity matrix S of all samples = {S (1) ,…,S (K)}. Two samples x i and x j The similarity of the k-th omics data Expressed as The specific calculation formula for structural similarity is as follows:

[0052]

[0053] in Represents x i The mean of Represents x j The mean of Represents x i The variance of Represents x j The variance of represents the covariance, c1 and c2 are constants. (k) represents the similarity matrix of the k-th omics data, It's S (k) The value of row i and column j in represents x i and x j similarity.

[0054] Then, the graph regularization formula for each omics is as follows:

[0055]

[0056] Where Tr() represents the trace of the matrix, L (k) is the Laplace transform of the graph, R (k) represents the graph regularization of the k-th omics data; u i is the i-th row of the shared matrix U, representing x i Linear integration characteristics; u j is the jth row of the shared matrix U, representing x j Linear integration characteristics.

[0057] The function of the graph regularization calculation formula (3) is to add the similarity information of the samples into the process of matrix decomposition to solve the problem that the traditional non-negative matrix decomposition method cannot mine the similarity information, thereby improving the final classification accuracy.

[0058] Step 3: Use variational autoencoders to obtain nonlinear integration features of multi-omics data samples.

[0059] The present invention uses variational autoencoders (VAE) to obtain nonlinear integration features of all samples. The specific process is: a multi-input and multi-output VAE model is constructed, and each set of input-output corresponds to a multi-omics data sample. The encoder f of the model encoder and decoder f decoder It can be expressed as the following formula:

[0060]

[0061]

[0062] where μ i and σ i Represent the mean vector and standard deviation vector respectively, and the goal is to make the embedding representation z i Obey Gaussian distribution z i The construction process uses the heavy parameter technique z i =μ i +σ i ∈, ∈ is a random variable sampled from the unit normal distribution. The μ obtained by VAE is i (1≤i≤M) as sample x i Nonlinear integration characteristics.

[0063] The encoder f of the VAE model encoder The input is a multi-omics sample The output is μ i and ρ i . Decoder f decoder The input is z i , the output is the reconstructed sample

[0064] In order to solve the problem that the learned nonlinear feature representation is meaningless and the variance in the feature space is overestimated, the present invention also uses the maximum mean difference (MMD) as the potential loss in the variational autoencoder.

[0065] Step 4: Concatenate the linear integration features, nonlinear integration features, and drug features of the multi-omics data sample to obtain a concatenated feature vector.

[0066] The linear integration feature u of the samples obtained by computer splicing i , nonlinear integration characteristics μ i and drug sample d, the splicing process is to transform the feature vector u i ,μ i and d are directly concatenated end to end to obtain the concatenated feature vector.

[0067] Step 5: Using the concatenated feature vector to train a drug sensitivity classifier.

[0068] The concatenated feature vectors are then used to train the classifier to achieve drug sensitivity classification. The training process uses a 5-fold cross-validation method for training and testing. The input of the drug sensitivity classifier is the feature vector u i ,μ iThe output is the feature vector of the concatenation of d and d. The output is the drug sensitivity category, which includes sensitive, resistant, and intermediate ranges. The 5-fold cross-validation method divides the dataset into five equally proportional, non-intersecting subsets. Four of these subsets are selected as the training set, and the remaining subset is used as the test set to obtain the classification results.

[0069] Step 6: Use the trained drug sensitivity classifier to perform drug sensitivity classification and obtain the classification results.

[0070] The trained drug sensitivity classifier is used to classify the drug sensitivity of biological sample cell lines, and the corresponding drug sensitivity category is obtained as the classification result. An example of the classification result is shown in Table 1 below:

[0071] Table 1

[0072] sample Prediction results (Multi-omics sample of cell line 1, drug 1) sensitive (Multi-omics samples of cell line 1, drug 2) resistance (Multi-omics samples of cell line 1, drug 3) sensitive

[0073] The results in Table 1 show that in cell line 1, Drug 1 and Drug 3 were predicted as "sensitive," while Drug 2 was predicted as "resistant." Therefore, Drug 1 and Drug 3 are effective in cell line 1. Therefore, Drug 1 and Drug 3 should be screened during drug screening. This shows that accurate drug sensitivity classification is beneficial for selecting promising drugs and assists in drug screening.

[0074] This paper proposes a drug sensitivity classification method based on multi-omics data integration. It extracts linear and nonlinear integration features from multi-omics data and then uses these features along with drug characteristics for classification. By integrating multi-omics data using linear and nonlinear integration features, the complementarity of these two features can be exploited to uncover potential information within the multi-omics data, helping to improve the accuracy of drug sensitivity classification.

[0075] Based on the method provided by the present invention, the present invention also provides a drug sensitivity classification system based on multi-omics data integration, comprising:

[0076] A sample collection module is used to obtain multi-omics data samples by sequencing cell lines and construct multiple omics matrices;

[0077] a linear integration module for decomposing the multiple omics matrices using joint non-negative matrix factorization, and adding graph regularization to the matrix decomposition by constructing structural similarity of the multi-omics data samples to obtain linear integration features of the multi-omics data samples;

[0078] Nonlinear integration module, used to obtain nonlinear integration features of multi-omics data samples using variational autoencoders;

[0079] A feature splicing module, configured to splice the linear integration features, nonlinear integration features, and drug features of the multi-omics data sample to obtain a spliced ​​feature vector;

[0080] a classifier training module, configured to train a drug sensitivity classifier using the concatenated feature vector;

[0081] The sensitivity classification module is used to perform drug sensitivity classification using a trained drug sensitivity classifier to obtain classification results.

[0082] Furthermore, the present invention provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other via the communication bus. The processor may invoke a computer program in the memory to execute the multi-omics data integration-based drug sensitivity classification method.

[0083] In addition, when the computer program in the above-mentioned memory is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk.

[0084] Furthermore, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed, can implement the drug sensitivity classification method based on multi-omics data integration.

[0085] Compared with existing multi-omics data integration and drug sensitivity classification methods, the method proposed in this paper demonstrates higher accuracy in drug sensitivity classification and better generalization across different datasets. This is because the method uses both linear and nonlinear integration features. The linear integration feature contains important information from all omics, eliminating noise in the original data; the nonlinear integration feature contains information about nonlinearities and complex relationships in multi-omics data. Leveraging the complementarity of these two features, the method mines potential information across multi-omics data, significantly improving the accuracy of drug sensitivity classification.

[0086] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0087] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A drug sensitivity classification method based on multi-omics data integration, characterized in that: include: Obtain multi-omics data samples by sequencing cell lines and construct multiple omics matrices; The method of obtaining multi-omics data samples by sequencing the cell lines and forming multiple omics matrices specifically includes: Obtaining multi-omics data samples by sequencing cell lines 1≤i≤M; M is the number of multi-omics data samples obtained; To form the multi-omics data sample x i The k-th omics data, 1≤k≤K; K is the multi-omics data sample x i The number of omics data included in the ; the omics data include gene expression and DNA methylation; The M multi-omics data samples obtained are used to form K omics matrices X = {X (1) ,…,X (K) }; where the k-th omics matrix Decomposing the multiple omics matrices using joint non-negative matrix factorization, and adding graph regularization to the matrix decomposition by constructing structural similarity of the multi-omics data samples to obtain linear integration features of the multi-omics data samples; The multiple omics matrices are decomposed using joint non-negative matrix factorization, and graph regularization is added to the matrix decomposition by constructing the structural similarity of the multi-omics data samples to obtain the linear integration features of the multi-omics data samples, specifically including: The K omics matrices X = {X (1) ,…,X (K) }Decomposed into a shared matrix U and K transformation matrices V (1) ,…,V (K) In the decomposition process, the structural similarity of multi-omics data samples is constructed to add graph regularization to the matrix decomposition, and the row vector u of the shared matrix U obtained by decomposition is i As a multi-omics data sample x i Linear integration characteristics; Use variational autoencoders to obtain nonlinear integration features of multi-omics data samples; splicing the linear integration features, the nonlinear integration features, and the drug features of the multi-omics data sample to obtain a spliced ​​feature vector; training a drug sensitivity classifier using the concatenated feature vector; The trained drug sensitivity classifier is used to perform drug sensitivity classification to obtain the classification results.

2. The drug sensitivity classification method based on multi-omics data integration according to claim 1, characterized in that The nonlinear integration features of multi-omics data samples obtained by using a variational autoencoder specifically include: Using the variational autoencoder encoder f encoder , using the formula Generate multi-omics data sample x i The corresponding mean vector μ i and the standard deviation vector σ i ; The obtained mean vector μ i As a multi-omics data sample x i Nonlinear integration characteristics.

3. The drug sensitivity classification method based on multi-omics data integration according to claim 2, characterized in that: The step of splicing the linear integration features, the nonlinear integration features, and the drug features of the multi-omics data sample to obtain a spliced ​​feature vector specifically includes: The multi-omics data sample x i The linear integration characteristic u i , nonlinear integration characteristics μ i The drug feature d is directly concatenated head to tail to obtain the concatenated feature vector.

4. The drug sensitivity classification method based on multi-omics data integration according to claim 3, characterized in that: The concatenated feature vector is used to train a drug sensitivity classifier, specifically including: The concatenated feature vector is used to train a drug sensitivity classifier, and the training process adopts a 5-fold cross-validation method for training and testing to obtain a trained drug sensitivity classifier; the input of the drug sensitivity classifier is the concatenated feature vector, and the output is the drug sensitivity category, which includes sensitive, resistant and intermediate areas.

5. A drug sensitivity classification system based on multi-omics data integration, characterized by: For implementing the drug sensitivity classification method based on multi-omics data integration according to any one of claims 1 to 4, the drug sensitivity classification system based on multi-omics data integration comprises: A sample collection module is used to obtain multi-omics data samples by sequencing cell lines and construct multiple omics matrices; a linear integration module for decomposing the multiple omics matrices using joint non-negative matrix factorization, and adding graph regularization to the matrix decomposition by constructing structural similarity of the multi-omics data samples to obtain linear integration features of the multi-omics data samples; Nonlinear integration module, used to obtain nonlinear integration features of multi-omics data samples using variational autoencoders; A feature splicing module, configured to splice the linear integration features, nonlinear integration features, and drug features of the multi-omics data sample to obtain a spliced ​​feature vector; a classifier training module, configured to train a drug sensitivity classifier using the concatenated feature vector; The sensitivity classification module is used to perform drug sensitivity classification using a trained drug sensitivity classifier to obtain classification results.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the drug sensitivity classification method based on multi-omics data integration according to any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the drug sensitivity classification method based on multi-omics data integration according to any one of claims 1 to 4 is implemented.

Images