Drug and pathway association prediction method and system based on supervised learning, and medium
By constructing the similarity matrix of drugs and pathways and splicing feature vectors, combining multiple supervised learning algorithms, the problems of high-dimensionality of data and insufficient generalization capabilities in drug-signal pathway correlation prediction are solved, achieving more efficient and accurate prediction effects.
Patent Information
- Application Number
- CN202411732017.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-03-07
AI Technical Summary
The prior art has problems in the high-dimensionality of data in the correlation prediction of drugs and signal pathways, resulting in complex computing, large resource consumption, and insufficient model generalization capabilities, and lack of unified analysis methods to integrate multimodal information.
By constructing the drug similarity matrix and the pathway similarity matrix, the drug pathway sample feature representation vector is obtained by splicing, the feature dimension is reduced, and the drug pathway correlation prediction model is constructed with supervised learning algorithms such as support vector machine, random forest, decision tree and AdaBoost.
It effectively reduces computational complexity and resource consumption, improves data processing efficiency, enhances the generalization ability and prediction accuracy of the model, provides a more compact and informative feature representation, and improves the efficiency and accuracy of drug pathway correlation prediction.
Smart Images
Figure CN119580871A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of bioinformatics, and in particular relates to a method, system and medium for predicting the association between drugs and pathways based on supervised learning. Background Art
[0002] Biological signaling pathways consist of a series of molecular interactions and signal transduction networks inside and outside cells, regulating key biological processes such as cell proliferation, differentiation, apoptosis and metabolism. The occurrence and development of many diseases, especially cancer, neurodegenerative diseases and metabolic disorders, are accompanied by abnormal activation or inhibition of specific signaling pathways. For example, overactivation of the PI3K / AKT pathway in cancer and dysfunction of the insulin signaling pathway in diabetes. Therefore, understanding how drugs affect these pathways is an important way to optimize drug therapy and improve drug efficacy and safety.
[0003] In traditional drug development, the selection of drug targets usually relies on known biological mechanisms and experimental verification. However, with the progress of high-throughput genomics, transcriptomics and proteomics technologies, the relationship between drugs and targets has become more complex. Drugs usually exert their therapeutic effects by affecting multiple targets or signaling pathways, and the interactions and network effects between these pathways often determine the efficacy and side effects of drugs. Take sorafenib (Sorafenib) for the treatment of hepatocellular carcinoma as an example. It is a multi-target anticancer drug. Sorafenib can inhibit RAF kinase in the RAF / MEK / ERK signaling pathway to prevent tumor cell proliferation, while targeting VEGFR and PDGFR to inhibit tumor angiogenesis. However, these signaling pathways do not exist in isolation, but are related to each other through complex networks. Therefore, while sorafenib effectively inhibits tumor growth, it may also cause side effects such as hypertension and skin toxicity by affecting non-target pathways. This fully illustrates the important role of pathway network effects in drug efficacy and safety. By predicting the association between drugs and specific pathways, researchers can understand the mechanism of action of drugs from a more systematic and global perspective, and improve the efficiency and success rate of drug development. In addition, drug-pathway association prediction is not only of great significance for new drug development, but also of great value for drug repositioning (i.e., using existing drugs for new indications). Many existing drugs may exert new therapeutic effects by regulating understudied pathways. By predicting the potential association between drugs and pathways, researchers can discover undiscovered indications for these drugs, thereby accelerating the clinical transformation process.
[0004] At present, many researchers have proposed and designed some computational models to predict the potential association between drugs and pathways. According to different technical methods, these models can be roughly divided into three categories: models based on Bayesian sparse factors, models based on matrix decomposition, and other types of machine learning models. In the Bayesian sparse factor method and matrix decomposition method, studies usually use gene expression data or drug sensitivity data of different cell lines to infer the potential connection between drugs and pathways. In the Bayesian sparse factor method, linear models are usually used to describe the relationship between drugs and pathways, and the collapsed Gibbs sampling algorithm is used for Bayesian reasoning. Representative models include iFad and FacPad. The matrix decomposition method transforms the problem into an optimization model through regularized optimization, obtains the feature matrix by solving the objective function, and then makes predictions. Common models include iPaD, L2, 1-iPaD, L1L2, 1-iPaD and IGMF. Bayesian sparse factor models usually describe the relationship between drugs and pathways based on linear models. This method shows limitations in capturing complex nonlinear associations. In addition, the model needs to rely on iterative sampling techniques (such as collapsed Gibbs sampling) for Bayesian inference, which has a high computational cost. Especially when facing large-scale, high-dimensional data, the efficiency is further reduced, and it is difficult to adapt to the needs of actual application scenarios. The matrix decomposition model uses the low-rank characteristics of the data matrix to compress high-dimensional data into low-dimensional representations to extract key features. However, this low-rank assumption may lead to the loss of some key information, especially when the data contains high noise, the prediction effect of the model may be significantly affected. In addition, the matrix decomposition method usually requires preset regularization constraints (such as sparsity or low rank), but these prior assumptions may not be completely consistent with the true distribution of the actual data, thereby limiting the generalization ability of the model. When the data matrix is too sparse, the reliability of the decomposition result will be significantly reduced, resulting in unstable prediction accuracy. Machine learning-based models integrate multiple information such as the chemical structure of the drug and pathway-related gene expression data as features for representation. These models are usually trained using classifiers and use training samples to predict the association between drugs and pathways. Common representative models include RGRF and LSA-PU-KNN. Its training usually relies on large-scale, high-quality datasets. However, in the task of drug-pathway association prediction, the datasets are often small and contain high-dimensional features (such as drug chemical descriptors, gene expression profiles, etc.), which makes it difficult to train the model. In this case, complex models (such as deep learning) are prone to overfitting problems, that is, the model performs well on the training set, but the generalization ability in the test set or actual application is insufficient.
[0005] More importantly, the current methods and models in the field of drug-pathway association prediction are still relatively limited. There is still a lack of a unified analysis method that can effectively integrate drug sensitivity data, gene expression data, and multimodal information of pathways to construct prediction models. This limits the further development of research in the field of drug-pathway association prediction and also hinders the widespread promotion of prediction models in practical biomedical applications. Summary of the invention
[0006] The present invention application provides a drug and pathway association prediction method, system, and medium based on supervised learning, aiming to partially or completely solve the defects of existing methods and models in the field of drug and pathway association prediction. The present invention application can obtain the association probability between each input drug and pathway sample pair, thereby identifying potential drug and pathway associations, and providing valuable information for drug development and biological research. In order to achieve the above purpose, the present invention application adopts the following technical solutions:
[0007] In the first aspect, the drug and pathway association prediction method based on supervised learning includes the following steps:
[0008] Step S100: obtaining a training sample set associated with a drug pathway;
[0009] Step S200: constructing multiple drug pathway sample feature representation vectors;
[0010] Step S300: constructing a drug pathway association prediction model based on a supervised learning training algorithm at least according to a training sample set associated with a drug pathway and a plurality of drug pathway sample feature representation vectors;
[0011] Step S400: Acquire the actual drug pathway sample feature representation vector, input it into the drug pathway association prediction model, and the drug pathway association prediction model predicts and outputs the drug and pathway association result.
[0012] Optionally, step S100 includes:
[0013] Step S101: In CCLE, 195 drug pathway positive samples are obtained, and the 195 drug pathway positive samples include 22 drugs and 58 pathways;
[0014] Step S102: The 195 drug pathway positive samples are classified into the positive sample set P, and 195 drug pathway samples are randomly selected from the remaining 1081 drug pathway samples and classified into the negative sample set N. The positive sample set P and the negative sample set N are used to form a training sample set associated with the drug pathway.
[0015] Optionally, step S200 includes:
[0016] Step S201: Calculate the similarity value S between drug i and drug j d (i,j), using the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d ;
[0017] Step S202: Calculate the similarity value S between path m and path n p (m,n), using the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p ;
[0018] Step S203: Based on the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[S d (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the path similarity matrix S p The m-th row vector of .
[0019] Optionally, in step S201, the similarity value S between drug i and drug j d (i,j) is:
[0020]
[0021] Where nc represents the number of cell lines in CCLE, nc = 480, Y(k,i) and Y(k,j) represent the sensitivity values of drug i and drug j in cell line k, respectively. and Respectively represent the average sensitivity values of drug i and drug j in all cell lines, k = 1, 2, ..., nc, i = 1, 2, ..., 22, j = 1, 2, ..., 22, i is a positive integer, j is a positive integer.
[0022] Optionally, in step S202, the similarity value S between path m and path n p (m,n) is:
[0023] S p (m,n)=exp(-α p ||L(m,:)-L(n,:)|| 2 )
[0024]
[0025] Where np represents the number of pathways in CCLE, np=58, exp is the exponential function, ||·|| is the bi-norm, L(m,:) and L(n,:) represent the mth and nth row vectors in the pathway gene association matrix L, respectively, α p is a parameter for controlling the kernel bandwidth, m=1, 2, ..., np, n=1, 2, ..., np, m is a positive integer, and n is a positive integer.
[0026] Optionally, step S300 includes:
[0027] Step S301: 195 drug pathway positive samples in the positive sample set P, 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway positive samples, and 195 labels 1 are associated in a one-to-one correspondence, and 195 drug pathway negative samples in the negative sample set, 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway negative samples, and 195 labels 0 are associated in a one-to-one correspondence to form drug pathway sample association data;
[0028] Step S302: Based on the supervised learning training algorithm, the Scikit-learn machine learning library is used to perform supervised learning training on the drug pathway sample association data, thereby obtaining a drug pathway association prediction model.
[0029] In a second aspect, a drug and pathway association prediction system based on supervised learning is used to implement any of the drug and pathway association prediction methods based on supervised learning described in the first aspect above, comprising:
[0030] A drug pathway sample feature representation vector construction unit, used to obtain multiple drug pathway sample feature representation vectors;
[0031] an acquisition unit, which acquires an actual drug pathway sample feature representation vector according to the multiple drug pathway sample feature representation vectors acquired by the drug pathway sample feature representation vector construction unit;
[0032] The prediction unit inputs the actual drug pathway sample feature representation vector into the drug pathway association prediction model, and the drug pathway association prediction model predicts and outputs the probability of association between drugs and pathways;
[0033] A display unit displays and outputs drug and pathway association results.
[0034] Optionally, the drug pathway sample feature representation vector construction unit includes: a drug similarity calculation unit, a pathway similarity calculation unit and a drug pathway sample feature representation vector construction unit;
[0035] The drug similarity calculation unit calculates the similarity value S between drug i and drug j d (i,j), and use the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d ;
[0036] The path similarity calculation unit calculates the similarity value S between path m and path n. p (m,n), and use the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p ;
[0037] The drug pathway sample feature representation vector construction unit is based on the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[S d (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the path similarity matrix S p The m-th row vector of .
[0038] Optionally, the drug and pathway association prediction system based on supervised learning also includes: a construction unit for constructing a drug-pathway association prediction model, the construction unit including: a sample set acquisition module for a training sample set for drug-pathway association; a supervised learning training module, which forms a drug-pathway association prediction model based on a supervised learning training algorithm, at least according to a training sample set for drug-pathway association and a plurality of drug pathway sample feature representation vectors.
[0039] In a third aspect, a computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the drug and pathway association prediction method based on supervised learning described in any one of the above-mentioned first aspects is executed.
[0040] (1) In the present invention, the drug similarity matrix S is proposed and calculated. d and the pathway similarity matrix S p , and according to the drug similarity matrix S d and the pathway similarity matrix S pBy concatenating drug similarity and pathway similarity vectors, a drug pathway sample feature representation vector Feature(i,m) is constructed. Instead of directly using drug sensitivity data or pathway gene association spectrum data, the feature dimension is greatly reduced, effectively avoiding the problems of complex calculations and high resource consumption caused by high-dimensional data, improving data processing efficiency, and having obvious dimensionality reduction advantages. The potential association information between drugs and pathways is still retained, building an accurate and efficient data foundation for subsequent drug pathway association prediction model training, so that the drug pathway association prediction model can be supervised learning training based on a more refined and valuable feature representation vector, improving the effect and accuracy of supervised learning training, and also providing a more compact and information-rich feature representation for supervised learning training of the drug pathway association prediction model.
[0041] (2) In the present invention application, supervised learning algorithms such as support vector machine, random forest, decision tree and AdaBoost were combined to complete the construction of four drug pathway association prediction models, namely DPASVM, DPARF, DPADT and DPAAda. The four drug pathway association prediction models, namely DPASVM, DPARF, DPADT and DPAAda, can all predict the potential association between drugs and pathways. The drug and pathway association prediction method based on supervised learning has good effectiveness and high feasibility.
[0042] (3) In the present invention, by conducting 100 five-fold cross validations, compared with the traditional single cross validation, the generalization ability and stability of the model can be more fully evaluated, the interference of accidental factors is reduced, and the performance of the four drug-pathway association prediction models, DPASVM, DPARF, DPADT, and DPAAda, under different data divisions can be more accurately reflected. The DPASVM pathway association prediction model has better performance in the drug and pathway association prediction task. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments are briefly introduced below. 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 creative work.
[0044] Figure 1 A schematic diagram of the process of the drug and pathway association prediction method based on supervised learning applied for by the present invention;
[0045] Figure 2 A schematic diagram of the construction principle of the drug pathway association prediction model applied for in the present invention;
[0046] Figure 3A schematic diagram of a drug and pathway association prediction system based on supervised learning applied for by the present invention;
[0047] Figure 4 This is a schematic diagram of the composition of the building blocks of the present invention.
[0048] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION
[0049] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments; based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0050] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside", "outside" and the like indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention. In order to make the purpose, technical solution and advantages of the present invention clearer, the implementation methods of the present invention will be further described in detail below in conjunction with the drawings.
[0051] Drug and pathway association prediction method based on supervised learning
[0052] like Figure 1 As shown, in the first aspect, a drug and pathway association prediction method based on supervised learning includes the following steps:
[0053] Step S100: obtaining a training sample set associated with a drug pathway;
[0054] Specifically, the steps S100 include:
[0055] Step S101: In the Cancer Cell Line Encyclopedia (CCLE), 195 positive samples of drug pathways are obtained. The 195 positive samples of drug pathways include 22 drugs and 58 pathways.
[0056] In the present invention, CCLE includes 195 known samples of drugs and pathways that are positively associated, that is, 195 drug pathway positive samples, 195 drug pathway positive samples include 22 drugs and 58 pathways, 22 drugs and 58 pathways can form 1276 drug pathway related samples, and the related samples can be represented by the drug pathway association matrix A∈R 22×58 To express.
[0057] Step S102: 195 drug pathway positive samples are classified into the positive sample set P, and 195 drug pathway samples are randomly selected from the remaining 1081 drug pathway samples and classified into the negative sample set N. The positive sample set P and the negative sample set N are used to form a training sample set associated with the drug pathway.
[0058] In the present invention application, in CCLE, 195 drug pathway associated samples are regarded as positive samples, which can be assigned label 1 and included in set P, that is, a positive sample set P containing label 1 can be formed; at the same time, after excluding the above-mentioned 195 known positive samples, a random selection strategy is adopted to randomly select 195 drug pathway samples from the remaining 1081 drug pathway samples. The 195 randomly selected drug pathway related samples have not been verified to be associated with the drug and the pathway. The 195 randomly selected drug pathway samples can be regarded as potential negative samples, which can be assigned label 0 and included in set N, that is, a negative sample set N containing label 0 can be formed, and a total of 195 drug pathway positive samples and 195 drug pathway negative samples are obtained. In this way, through a specific selection strategy, the positive sample set P and the negative sample set N are determined respectively, each containing 195 samples of equal number, which effectively ensures the balance of positive and negative samples in subsequent training and avoids model bias due to sample imbalance; at the same time, the positive samples are determined based on known related data, and the negative samples are reasonably selected and potentially irrelevant, making the data used for training more representative, thereby improving the model training effect and the accuracy of drug pathway association judgment.
[0059] Step S200: constructing multiple drug pathway sample feature representation vectors;
[0060] In some embodiments, for 22 drugs, sensitivity data of 480 cell lines can be obtained, and the drug-cell line sensitivity association matrix Y, Y∈R 22×480 ,The drug cell line sensitivity association matrix Y records the association information between these drugs and cell line ,sensitivity. 58 pathways can obtain the association data of 1802 genes corresponding to them. ,The pathway gene association matrix L,L∈R 58×1802 ,The pathway-gene association matrix L records the association information between these pathways and genes.
[0061] In some embodiments, in the process of constructing a drug and pathway association data set, it is necessary to obtain and collect the sensitivity data of the drug in the cell line and the association spectrum data of the pathway and the gene. If the sensitivity value of the drug in different cell lines is used directly and alone as the feature of the drug, the feature representation dimension of each drug will reach 480 dimensions, or the association spectrum of the pathway and the gene is used directly and alone as the feature of the pathway, then the feature representation dimension of each pathway will reach 1802 dimensions. However, due to the limited number, time, and cost of drug and pathway association prediction training samples, higher-dimensional or high-dimensional feature representations of 480 or 1802 dimensions will result in excessive sample feature dimensions, increased difficulty in sample training in the drug and pathway association prediction model, and the risk of overfitting. Accordingly, the present invention application proposes a method based on the splicing of drug similarity and pathway similarity vectors for constructing a drug pathway sample feature representation vector. Specifically, step S200 includes:
[0062] Step S201: Calculate the similarity value S between drug i and drug j d (i,j), using the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d ;
[0063] In some embodiments, the similarity value S between drug i and drug j can be calculated according to the Pearson correlation coefficient calculation method and the drug cell sensitivity association matrix Y: d (i,j), the specific calculation method is as follows:
[0064]
[0065] Where nc represents the number of cell lines in CCLE, nc = 480, Y(k,i) and Y(k,j) represent the sensitivity values of drug i and drug j in cell line k, respectively. and Respectively represent the average sensitivity values of drug i and drug j in all cell lines, k = 1, 2, ..., nc, i = 1, 2, ..., 22, j = 1, 2, ..., 22, i is a positive integer, j is a positive integer.
[0066] Thus, the similarity value S between drug i and drug j is d (i, j), each drug includes 22 similarity values, and 22 drugs will include 484 similarity values. Then, 484 similarity values can construct the drug similarity matrix S between all drugs. d Each drug includes 22 similarity values that can form a drug similarity matrix S dThe row vector of , which can then fully record the similarity information between all drugs. In this way, the drug similarity matrix S is constructed by the similarity values between two drugs. d It can cover all similarity associations between 22 drugs without missing any combination information, provide a panoramic perspective for drug analysis, integrate scattered similarity values into a matrix form, and enable the similarity information between drugs to be recorded in an orderly and centralized manner, facilitating data storage, management and subsequent complex analytical operations, greatly improving the efficiency of subsequent drug research.
[0067] Step S202: Calculate the similarity value S between path m and path n p (m,n), using the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p ;
[0068] In some embodiments, the similarity value S between pathway m and pathway n can be calculated based on the Gaussian spectral kernel method and the pathway gene association matrix L p (m,n), the specific calculation method is as follows:
[0069] S p (m,n)=exp(-α p ||L(m,:)-L(n,:)|| 2 )
[0070]
[0071] Where np represents the number of pathways in CCLE, np=58, exp is the exponential function, ||·|| is the bi-norm, L(m,:) and L(n,:) represent the mth and nth row vectors in the pathway gene association matrix L, respectively, α p is a parameter for controlling kernel bandwidth, m=1, 2, ..., np, n=1, 2, ..., np, m is a positive integer, and n is a positive integer. It should be noted that in the present invention, the association spectrum between pathway m and gene represents the association and association distribution between pathway m and gene, which can be represented by a binary vector L(m,:), which represents the row vector of the mth row of the pathway gene association matrix L corresponding to pathway m.
[0072] Thus, the similarity value S between path m and path n is p (m,n), each pathway includes 58 similarity values, and 58 drugs will include 3364 similarity values. Then, 3364 similarity values can construct the pathway similarity matrix S between all pathways. p , each path includes 58 similarity values that can form the path similarity matrix Sp The row vector of , which can then fully record the similarity information between all pathways. In this way, the matrix S is constructed by calculating the similarity values between pathways p It comprehensively covers the similar relationships between 58 pathways without any information loss, has high integrity and good integration, and organizes a large number of scattered similarity values into a matrix, making the pathway similarity information clear and easy to centrally manage and view, which is conducive to the efficient implementation of subsequent in-depth research and analysis based on pathway similarity.
[0073] Step S203: Based on the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[S d (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the path similarity matrix S p The m-th row vector of .
[0074] In some embodiments, for any drug i, the similarity characteristics of drug i can be expressed as the drug similarity matrix S d The i-th row vector S d (i,:), for any path m, the similarity feature of path m can be expressed as the path similarity matrix S p The mth row vector S p On this basis, for any drug i and any pathway m together to form a drug pathway sample F, the drug pathway sample feature representation vector of drug pathway sample F can be obtained by the drug i similarity feature S d (i,:) and the similarity feature S of pathway m p (m,:) is concatenated, that is, the drug pathway sample feature representation vector is Feature(i,m). The specific calculation formula is as follows:
[0075] Feature(i,m)=[S d (i,:),S p (m,:)]
[0076] Obviously, the dimension of the concatenated feature vector is significantly lower than the feature dimension of directly using the original sensitivity data and gene association spectrum data. The feature vector generated in this way effectively reduces the dimension of the original feature, greatly reduces the feature dimension, effectively avoids the problems of complex calculations and high resource consumption caused by high-dimensional data, improves data processing efficiency, and has obvious dimensionality reduction advantages. At the same time, although the dimension is reduced, the potential association between drugs and pathways can be accurately retained, but the potential association information between drugs and pathways is still retained to ensure that key information is not lost, and an accurate and efficient data foundation is built for the subsequent training of drug pathway association prediction models, so that the drug pathway association prediction model can be based on a more refined and valuable feature representation vector for supervised learning training, improve the effect and accuracy of supervised learning training, and provide a more compact and information-rich feature representation for supervised learning training of drug pathway association prediction models.
[0077] Step S300: constructing a drug pathway association prediction model based on a supervised learning training algorithm at least according to a training sample set associated with a drug pathway and a plurality of drug pathway sample feature representation vectors;
[0078] Step S301: 195 drug pathway positive samples in the positive sample set P, 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway positive samples, and 195 labels 1 are associated in a one-to-one correspondence, and 195 drug pathway negative samples in the negative sample set, 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway negative samples, and 195 labels 0 are associated in a one-to-one correspondence to form drug pathway sample association data;
[0079] Step S302: Based on the supervised learning training algorithm, the Scikit-learn machine learning library is used to perform supervised learning training on the drug pathway sample association data, thereby obtaining a drug pathway association prediction model.
[0080] like Figure 2 As shown, in the present invention application, firstly, by associating positive samples, negative samples and their corresponding feature vectors and labels one by one, the sample data is organized in order, providing a complete and accurate information source basis for the training of the drug pathway association prediction model; in addition, the Scikit-learn learning library is powerful and can quickly perform supervised learning training on the associated data, and accordingly, the construction of the drug pathway association prediction model can be completed efficiently and conveniently, saving the development time and energy of the drug pathway association prediction model. The constructed drug pathway association prediction model is highly practical and can efficiently output the predicted association of drug pathway samples, which is helpful for screening suitable or potential drug and pathway combinations in drug research and development, accelerating the drug research and development process, and improving the success rate of drug research and development.
[0081] Step S400: Acquire the actual drug pathway sample feature representation vector, input it into the drug pathway association prediction model, and the drug pathway association prediction model predicts and outputs the drug and pathway association result.
[0082] In the present application, according to multiple drug pathway sample feature representation vectors, one of the drug pathway sample feature representation vectors can be used as the actual drug pathway sample feature representation vector, and the actual drug pathway sample feature representation vector is input into the drug pathway association prediction model that has been trained based on supervised learning. Through the prediction of unknown drug pathway samples by the drug pathway association prediction model, the drug and pathway association results between the input drugs and pathways can be obtained. The drug and pathway association results may include the association probability value between the drug and the pathway or whether the association conclusion (including association or non-association), so that potential drug and pathway associations can be identified. Therefore, the database formed by these association probability values or association probability value sets or whether the association conclusion or whether the association conclusion is formed can be used as the key credential for in-depth analysis and verification of the association between drugs and pathways, and provide valuable reference information and data support for the drug development process and research and exploration in the field of biology.
[0083] Examples of drug and pathway association prediction methods based on supervised learning
[0084] In the present application, four common supervised learning training algorithms are adopted: Support Vector Machine (SVM), Random Forest, Decision Tree and AdaBoost (Adaptive Boosting). The previously constructed drug pathway association training sample set and the drug pathway sample feature representation vector and label corresponding to each sample are used to train the model based on Python programming language and Scikit-learn machine learning library, and a drug pathway association prediction model DPASVM (Drug-Pathway Association prediction based on Support Vector Machine, hereinafter referred to as DPASVM) based on support vector machine, a drug pathway association prediction model DPARF (Drug-Pathway Association prediction based on Random Forest, hereinafter referred to as DPARF) based on random forest, a drug pathway association prediction model DPADT (Drug-Pathway Association prediction based on Decision Tree, hereinafter referred to as DPADT) based on decision tree and a drug pathway association prediction model DPAAda (Drug-Pathway Association predictionbased on AdaBoost, hereinafter referred to as DPAAda) based on AdaBoost are respectively constructed.
[0085] The four drug-pathway association prediction models, DPASVM, DPARF, DPADT, and DPAAda, are all based on the Scikit-learn machine learning library in Python, and are constructed using SVC (Support Vector Machine), Random Forest Classifier, Decision Tree Classifier, and AdaBoost Classifier (AdaBoost) modules. The machine learning algorithms in the four models, DPASVM, DPARF, DPADT, and DPAAda, have their own characteristics, and can combine the advantages of specific algorithms to explore the potential associations between drugs and pathways from different angles. They can also improve classification performance by optimizing hyperparameters, ensuring the model's efficient mining capabilities and prediction performance for the potential associations between drugs and pathways.
[0086] After the construction of the above four drug-pathway association prediction models, DPASVM, DPARF, DPADT, and DPAAda, the feature vectors of unknown drug-pathway sample pairs are input into each drug-pathway association prediction model that has been trained based on supervised learning. Through the prediction of the drug-pathway association prediction model, the association probability between each input drug and pathway can be obtained, thereby identifying potential drug-pathway associations. These association probabilities will serve as an important basis for further analysis and verification of the relationship between drugs and pathways, providing valuable information for drug development and biological research.
[0087] In the present invention application, a 5-fold cross-validation method is used to evaluate the prediction performance of the proposed four drug pathway association prediction models, namely DPASVM, DPARF, DPADT, and DPAAda. In the 5-fold cross-validation, the data set will be randomly divided into five subsets, one of which will be selected as the test set each time, and the other four subsets will be used as training sets. This process will be repeated five times to ensure that each subset can participate in the model evaluation as a test set. Finally, the average of the five validation results will be calculated to obtain a comprehensive performance evaluation of the model.
[0088] In the evaluation process, AUC (Area Under the Curve) is used as the main metric. AUC can comprehensively consider the performance of the model under different classification thresholds, reflecting the model's ability to classify positive and negative samples, and is not affected by class imbalance. In this way, we can comprehensively and stably evaluate the performance of the proposed DPASVM, DPARF, DPADT, and DPAAda drug-pathway association prediction models in the drug-pathway association prediction task.
[0089] From the results in Table 1, it can be seen that the average AUC values of the drug-pathway association prediction model based on the supervised learning training algorithm constructed by the present invention are all over 0.8, which fully verifies the effectiveness of the drug-pathway association prediction method based on supervised learning proposed by the present invention, and also shows that the method of constructing sample feature vectors by splicing similarity has high feasibility, and proves the application value of support vector machine, random forest, decision tree and AdaBoost algorithm in the problem of drug-pathway association prediction. Among them, the support vector machine performs best in AUC and stability, fully demonstrating its powerful nonlinear feature extraction ability and superiority in the case of small samples.
[0090] Table 1 Five-fold cross-validation results of four drug pathway association prediction models: DPASVM, DPARF, DPADT, and DPAAda
[0091] Drug pathway association prediction model Average AUC value Standard Deviation DPASVM 0.8800 0.0069 DPARF 0.8635 0.0088 DPADT 0.8057 0.0147 DPAAda 0.8448 0.0094
[0092] Drug and pathway association prediction system based on supervised learning
[0093] like Figure 3 As shown, in the second aspect, a drug and pathway association prediction system based on supervised learning is used to implement any of the drug and pathway association prediction methods based on supervised learning described in the first aspect above, including:
[0094] It should be noted that the drug and pathway association prediction system based on supervised learning applied in the present invention is used to implement any drug and pathway association prediction method based on supervised learning in the first aspect, and accordingly also includes: all the recorded technical problems, technical solutions and technical effects of any drug and pathway association prediction method based on supervised learning in the first aspect, which will not be repeated in the present invention application.
[0095] A drug pathway sample feature representation vector construction unit 100 is used to obtain a plurality of drug pathway sample feature representation vectors;
[0096] An acquisition unit 200 acquires an actual drug pathway sample feature representation vector according to the multiple drug pathway sample feature representation vectors acquired by the drug pathway sample feature representation vector construction unit 100;
[0097] The prediction unit 300 inputs the actual drug pathway sample feature representation vector into the drug pathway association prediction model, and the drug pathway association prediction model predicts and outputs the probability of association between the drug and the pathway;
[0098] The display unit 400 displays and outputs the drug and pathway association results.
[0099] Optionally, the drug pathway sample feature representation vector construction unit 100 includes: a drug similarity calculation unit, a pathway similarity calculation unit and a drug pathway sample feature representation vector construction unit; the drug similarity calculation unit calculates the similarity value S between drug i and drug j d (i,j), and use the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d ; The path similarity calculation unit calculates the similarity value S between path m and path n p (m,n), and use the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p ; Drug pathway sample feature representation vector construction unit, according to the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[Sd (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the path similarity matrix S p The m-th row vector of .
[0100] In some embodiments, the drug similarity calculation unit calculates the similarity value S between drug i and drug j. d (i,j), and use the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d Specifically, the similarity value S between drug i and drug j can be calculated based on the Pearson correlation coefficient calculation method and the drug cell sensitivity association matrix Y. d (i,j), the specific calculation method is as follows:
[0101]
[0102] Where nc represents the number of cell lines in CCLE, nc = 480, Y(k,i) and Y(k,j) represent the sensitivity values of drug i and drug j in cell line k, respectively. and Respectively represent the average sensitivity values of drug i and drug j in all cell lines, k = 1, 2, ..., nc, i = 1, 2, ..., 22, j = 1, 2, ..., 22, i is a positive integer, j is a positive integer.
[0103] In some embodiments, the path similarity calculation unit calculates the similarity value S between path m and path n. p (m,n), and use the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p Specifically, the similarity value S between pathway m and pathway n can be calculated based on the Gaussian spectral kernel method and the pathway gene association matrix L. p (m,n), the specific calculation method is as follows:
[0104] S p (m,n)=exp(-α p ||L(m,:)-L(n,:)|| 2 )
[0105]
[0106] Where np represents the number of pathways in CCLE, np=58, exp is the exponential function, ||·|| is the bi-norm, L(m,:) and L(n,:) represent the mth and nth row vectors in the pathway gene association matrix L, respectively, α p is a parameter for controlling the kernel bandwidth, m=1, 2, ..., np, n=1, 2, ..., np, m is a positive integer, and n is a positive integer.
[0107] In some embodiments, the drug pathway sample feature representation vector construction unit is based on the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[S d (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the pathway similarity matrix S p The m-th row vector of .
[0108] Specifically, for any drug i, the similarity characteristics of drug i can be expressed as the drug similarity matrix S d The i-th row vector S d (i,:), for any path m, the similarity feature of path m can be expressed as the path similarity matrix S p The mth row vector S p On this basis, for a drug pathway sample F formed by any drug i and any pathway m, the drug pathway sample feature representation vector of drug pathway sample F can be expressed as the similarity feature S of drug i. d (i,:) and the similarity feature S of pathway m p (m,:) is concatenated, that is, the drug pathway sample feature representation vector is Feature(i,m). The specific calculation formula is as follows:
[0109] Feature(i,m)=[S d (i,:),S p (m,:)]
[0110] On this basis, the acquisition unit 200 can select one of the drug pathway sample feature representation vectors as the actual drug pathway sample feature representation vector based on the multiple drug pathway sample feature representation vectors acquired by the drug pathway sample feature representation vector construction unit 100 .
[0111] Alternatively, if Figure 4As shown, the drug and pathway association prediction system based on supervised learning also includes: a construction unit 500, which is used to construct a drug pathway association prediction model, and the construction unit 500 includes: a sample set acquisition module 501, which is used for a training sample set of drug pathway association; a supervised learning training module 502, which is based on at least the training sample set of drug pathway association and the drug pathway sample feature representation vector in the construction unit 100, and based on the supervised learning training algorithm, learns and trains to form a drug pathway association prediction model.
[0112] In some embodiments, the sample set acquisition module 501 is used for a training sample set associated with a drug pathway. Specifically, the sample set acquisition module 501 includes: in CCLE, obtaining 195 positive samples of drug pathways, the 195 positive samples of drug pathways include 22 drugs and 58 pathways; classifying the 195 positive samples of drug pathways into a positive sample set P, randomly selecting 195 drug pathway samples from the remaining 1081 drug pathway samples, and classifying them into a negative sample set N, and using the positive sample set P and the negative sample set N to form a training sample set associated with a drug pathway.
[0113] In some embodiments, the supervised learning training module 502, based on at least the training sample set associated with the drug pathway and the multiple drug pathway sample feature representation vectors in the drug pathway sample feature representation vector construction unit 100, learns and trains to form a drug pathway association prediction model based on the supervised learning training algorithm. Specifically, the supervised learning training module 403 includes: associating the 195 drug pathway positive samples in the positive sample set P, the 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway positive samples, and the 195 labels 1 in a one-to-one correspondence, associating the 195 drug pathway negative samples in the negative sample set, the 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway negative samples, and the 195 labels 0 in a one-to-one correspondence, forming drug pathway sample association data, and based on the supervised learning training algorithm, using the Scikit-learn machine learning library to perform supervised learning training on the drug pathway sample association data, thereby obtaining a drug pathway association prediction model.
[0114] Therefore, in the present application, firstly, the precise correspondence between drug pathway positive samples, drug pathway negative samples, drug pathway sample feature representation vectors and labels constructs structured drug pathway sample association basic data, providing a solid data foundation for drug pathway association prediction model training; in addition, based on the supervised learning training algorithm, supervised learning training is performed with the help of Scikit-learn's powerful machine learning library, which can efficiently utilize its rich algorithm tools and optimization strategies to quickly converge to obtain a stable and high-performance drug pathway association prediction model, which can accurately predict the association between drugs and pathways based on the input drug pathway sample feature representation vector, and effectively promote the progress and decision-making of drug research and development and related biological research.
[0115] In some embodiments, the display unit 400 may adopt a display screen, and the drug and pathway association results displayed by the display unit 400 may include the association probability value between the drug and the pathway or the conclusion on whether the drug is associated (including association or non-association). The database formed by these association probability values or sets of association probability values or the conclusion on whether the drug is associated or not or the database formed by the conclusion on whether the drug is associated can be used as the key credential for in-depth analysis and verification of the association between the drug and the pathway, and provide extremely valuable reference information and data support for the drug development process and research and exploration in the field of biology.
[0116] Computer readable storage medium
[0117] In a third aspect, the present invention application provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed on a computer, the drug and pathway association prediction method based on supervised learning as described in any one of the first aspects is executed.
[0118] Drug and pathway association prediction device based on supervised learning
[0119] In a fourth aspect, the present invention application provides a drug and pathway association prediction device based on supervised learning, comprising a memory and a processor that are communicatively connected, wherein the memory is used to store a computer program, and the processor is used to read the computer program and execute the drug and pathway association prediction method based on supervised learning as described in any one of the first aspects.
[0120] In some embodiments, the drug and pathway association prediction device based on supervised learning includes: a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps in any of the drug and pathway association prediction methods based on supervised learning in the first aspect are implemented.
[0121] Those skilled in the art will understand that the schematic diagram is merely an example of a terminal device and does not constitute a limitation on the terminal device. The terminal device may include more or fewer components than shown in the diagram, or a combination of certain components, or different components. For example, the terminal device may also include input and output devices, network access devices, buses, etc.
[0122] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, and uses various interfaces and lines to connect various parts of the entire terminal device.
[0123] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the terminal device by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.
[0124] Wherein, if the module / unit integrated in the terminal device 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 such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.
[0125] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.
[0126] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A drug and pathway association prediction method based on supervised learning, characterized in that: The following steps are involved: Step S100: obtaining a training sample set associated with a drug pathway; Step S200: constructing multiple drug pathway sample feature representation vectors; Step S300: constructing a drug pathway association prediction model based on a supervised learning training algorithm at least according to a training sample set associated with a drug pathway and a plurality of drug pathway sample feature representation vectors; Step S400: Acquire the actual drug pathway sample feature representation vector, input it into the drug pathway association prediction model, and the drug pathway association prediction model predicts and outputs the drug and pathway association result.
2. The drug and pathway association prediction method based on supervised learning according to claim 1, characterized in that: The steps S100 include: Step S101: In CCLE, 195 drug pathway positive samples are obtained, and the 195 drug pathway positive samples include 22 drugs and 58 pathways; Step S102: 195 drug pathway positive samples are classified into the positive sample set P, and 195 drug pathway samples are randomly selected from the remaining 1081 drug pathway samples and classified into the negative sample set N. The positive sample set P and the negative sample set N are used to form a training sample set associated with the drug pathway.
3. The drug and pathway association prediction method based on supervised learning according to claim 2, characterized in that: Step S200 includes: Step S201: Calculate the similarity value S between drug i and drug j d (i,j), using the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d ; Step S202: Calculate the similarity value S between path m and path n p (m,n), using the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p ; Step S203: Based on the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[S d (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the path similarity matrix S p The m-th row vector of .
4. According to the supervised learning-based drug and pathway association prediction method of claim 1, in step S201, the similarity value S between drug i and drug j is d (i,j) is: Where nc represents the number of cell lines in CCLE, nc = 480, Y(k,i) and Y(k,j) represent the sensitivity values of drug i and drug j in cell line k, respectively. and Respectively represent the average sensitivity values of drug i and drug j in all cell lines, k = 1, 2, ..., nc, i = 1, 2, ..., 22, j = 1, 2, ..., 22, i is a positive integer, j is a positive integer.
5. The drug and pathway association prediction method based on supervised learning according to claim 4, characterized in that: In step S202, the similarity value S between path m and path n is p (m,n) is: S p (m,n)=exp(-α p ||L(m,:)-L(n,:)|| 2 ) Where np represents the number of pathways in CCLE, np=58, exp is the exponential function, ||·| is the bi-norm, L(m,:) and L(n,:) represent the mth and nth row vectors in the pathway gene association matrix L, respectively, α p is a parameter for controlling the kernel bandwidth, m=1, 2, ..., np, n=1, 2, ..., np, m is a positive integer, and n is a positive integer.
6. The drug and pathway association prediction method based on supervised learning according to claim 2, characterized in that: Step S300 includes: Step S301: 195 drug pathway positive samples in the positive sample set P, 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway positive samples, and 195 labels 1 are associated in a one-to-one correspondence, and 195 drug pathway negative samples in the negative sample set, 195 drug pathway sample feature representation vectors corresponding to the 195 drug pathway negative samples, and 195 labels 0 are associated in a one-to-one correspondence to form drug pathway sample association data; Step S302: Based on the supervised learning training algorithm, the Scikit-learn machine learning library is used to perform supervised learning training on the drug pathway sample association data, thereby obtaining a drug pathway association prediction model.
7. A drug and pathway association prediction system based on supervised learning, used to implement the drug and pathway association prediction method based on supervised learning according to any one of claims 1 to 6, characterized in that: include: A drug pathway sample feature representation vector construction unit, used to obtain multiple drug pathway sample feature representation vectors; an acquisition unit, which acquires an actual drug pathway sample feature representation vector according to the multiple drug pathway sample feature representation vectors acquired by the drug pathway sample feature representation vector construction unit; The prediction unit inputs the actual drug pathway sample feature representation vector into the drug pathway association prediction model, and the drug pathway association prediction model predicts and outputs the probability of association between drugs and pathways; A display unit displays and outputs drug and pathway association results.
8. The drug and pathway association prediction system based on supervised learning according to claim 7, characterized in that: The drug pathway sample feature representation vector construction unit includes: a drug similarity calculation unit, a pathway similarity calculation unit and a drug pathway sample feature representation vector construction unit; The drug similarity calculation unit calculates the similarity value S between drug i and drug j d (i,j), and use the similarity value S d (i,j) Construct the drug similarity matrix S between all drugs d ; The path similarity calculation unit calculates the similarity value S between path m and path n. p (m,n), and use the similarity value S p (m,n) construct the pathway similarity matrix S between all pathways p ; The drug pathway sample feature representation vector construction unit is based on the drug similarity matrix S d and the pathway similarity matrix S p , obtain the drug pathway sample feature representation vector Feature(i,m), Feature(i,m)=[S d (i,:),S p (m,:)], where S d (i,:) is the drug similarity matrix S d The i-th row vector, S p (m,:) is the path similarity matrix S p The m-th row vector of .
9. The drug and pathway association prediction system based on supervised learning according to claim 7, characterized in that: The drug and pathway association prediction system based on supervised learning also includes: a construction unit, which is used to construct a drug-pathway association prediction model, and the construction unit includes: a sample set acquisition module, which is used for a training sample set of drug-pathway association; a supervised learning training module, which forms a drug-pathway association prediction model based on a supervised learning training algorithm, based on at least a training sample set of drug-pathway association and a plurality of drug pathway sample feature representation vectors.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a computer, the drug and pathway association prediction method based on supervised learning described in any one of claims 1 to 6 is executed.