Drug recommendation method and system for drug effect prediction and / or personalized medication guidance

Abstract

The invention provides a drug recommendation method for drug effect prediction and / or personalized medication guidance. The drug recommendation method is based on a multi-dimensional differential expression map reference database. According to the invention, the association between the gene and a drug action mechanism or a disease phenotype is inferred through multi-dimensional information of the gene and functional modules including a GO function, a signal channel, transcriptional regulation, protein-protein interaction and the like; and meanwhile, a model integrated statistical method is benefited, and more accurate drug effect prediction and personalized medication guidance are finally realized compared with similar work. The method has a wide application prospect.

Description

Technical Field

[0001] The present invention belongs to computational biology-assisted drug development technology, relates to the field of biomedicine, and in particular to a drug recommendation method and system for drug efficacy prediction and / or personalized medication guidance, which is based on multi-dimensional differential expression maps and uses integrated statistical methods to predict drug efficacy and / or guide personalized medication. Background Art

[0002] With the development of sequencing technology and computational biology, the use of gene expression map data, especially drug perturbation expression map data, to predict drug efficacy and guide personalized medication has become a hot topic in the biomedical field in recent years. At present, data resources such as the Connectivity Map CMap (The Connectivity Map), the Library of Integrated Network-Based Cellular Signatures (LINCS) and the Genomics of Drug Sensitivity in Cancer (GDSC) have been released in the field. They contain gene expression maps before and after cells are perturbed in various ways (such as knocking out or overexpressing genes, compound treatment, etc., mainly compound perturbation) and information on the sensitivity of tumor cell lines to drugs.

[0003] Systematic collection and analysis of genome-wide compound-perturbation gene expression data from human cell lines or other samples can be used to explore drug similarities or new uses, discover drug repurposing opportunities, or guide personalized medicine. The phenotype prediction based on gene function modules is more robust than that based on genes alone [Fernández-Torras, A., Duran-Frigola, M. & Aloy, P. Encircling the regions of the pharmacogenomic landscape that determine drug response. Genome Med. 2019.], but most of the currently available drug efficacy prediction tools use methods based only on the gene level, such as Dr Insight, signatureSearch, and models based on deep neural networks [Chan J, Wang X, Turner JA, Baldwin NE, Gu J. Breaking the paradigm: Dr Insight empowers signature-free, enhanced drug repurposing. Bioinformatics. 2019; Duan Y, Evans DS, Miller RA, Schork NJ, Cummings SR, Girke T. signatureSearch: environment for gene expression signature searching and functional interpretation. Nucleic Acids Res. 2020; Jia P, Hu R, Pei G, Dai Y, Wang YY, Zhao Z. Deep generative neural network for accurate drug response imputation. Nat Commun. 2021.】, these methods have failed to make full use of gene function information, and only use the single-dimensional expression characteristics of the gene level for prediction, ignoring the association between different genes in the biological system, such as functional redundancy, complementarity, emergence and other characteristics, thereby greatly reducing the usable value of the above data, resulting in false positives and false negatives in the prediction results; at the same time, it is impossible to give the association between gene expression and drug phenotype from higher dimensions such as signal pathways and transcriptional regulation, reducing the interpretability of the prediction results. In fact, the functional significance of genes needs to be considered in the biomolecular network, and genes achieve specific functions through interactions and collaborations with other genes.If only a single gene dimension is considered, genes with relatively small changes but significant synergy in specific functional modules tend to be ignored. Therefore, expanding transcriptomics data from genes to functional gene set levels helps to interpret and utilize data from a dimension that is more biologically meaningful and more in line with the research ideas of systems biology. However, in the public work on predicting drug efficacy based on expression maps, there is almost no expansion of gene-level data to the functional gene set level. The present invention combines multi-dimensional information such as genes and gene function modules, and benefits from the statistical method of model integration, so the prediction results are more robust and extensible. Summary of the invention

[0004] The present invention overcomes the limitations of the prior art, and in order to achieve more accurate drug efficacy prediction and personalized medication guidance, an innovative multi-dimensional differential expression map data mining strategy is proposed and a corresponding method and system are constructed, making full use of the association between multi-dimensional information such as genes, GO functions, signal pathways, transcriptional regulation, protein-protein interactions, and drug phenotypes, not only from the gene level, but also from the aspects of GO functions, signal pathways, transcriptional regulation, protein-protein interactions, etc., to conduct a multi-dimensional comprehensive system analysis, and use integrated statistical methods to achieve more accurate drug efficacy prediction and personalized medication guidance. The present invention is based on a solid theoretical foundation, breaks through the limitations and difficulties of the prior art, is easy to implement, and realizes the application of promoting accurate prediction of drug efficacy and guiding personalized medication regimens.

[0005] The present invention proposes a drug recommendation method based on a multidimensional differential expression map for predicting drug efficacy and guiding personalized drug use. Based on the multidimensional differential expression map and a multidimensional differential expression map reference database constructed based on the multidimensional differential expression map, drug recommendation ranking results corresponding to multiple dimensions are obtained through a similarity measurement method; finally, an integrated statistical method is used to merge the drug recommendation ranking results under different dimensions to obtain a comprehensive ranking result, which can be used to accurately predict drug efficacy (including but not limited to drug similarity prediction, drug function prediction, etc.) and guide personalized drug use.

[0006] The present invention proposes a drug recommendation method based on a multidimensional differential expression map reference database, which infers the association with the drug action mechanism or disease phenotype through multidimensional information of genes and functional modules including GO function, signal pathway, transcriptional regulation, protein-protein interaction, etc., and then uses statistical methods of model integration to achieve drug recommendation; the drug recommendation includes efficacy prediction and / or personalized medication guidance, and the drug includes compound.

[0007] In the present invention, the multidimensional differential expression map refers to a differential expression map in multiple dimensions such as genes, GO functions, signal pathways, transcriptional regulation, protein-protein interactions, etc., including: a perturbation multidimensional differential expression map and a drug sensitivity multidimensional differential expression map.

[0008] In the present invention, the multidimensional differential expression map reference database is a database obtained by summarizing differential expression maps constructed based on different dimensions, and a perturbation reference database or a drug sensitivity reference database is obtained corresponding to the perturbation multidimensional differential expression map or the drug sensitivity multidimensional differential expression map, respectively.

[0009] In the present invention, the drug efficacy prediction refers to drug similarity prediction, drug function prediction, etc., which can be used for drug reuse or drug mechanism discovery. This function is achieved based on the comparison of the similarity between the unknown perturbation differential expression data (input data) and the known perturbation differential expression data (reference data). The unknown and known here indicate whether the mechanism of action or effect of the perturbation is clear, such as the input data is a compound with an unknown mechanism or effect; while the mechanism of action or effect of the comparative perturbation is known. If the differential expression caused by different perturbations is similar, then different perturbations may have similar mechanisms of action or effects, so that the mechanism of action or effect of the known perturbation can be migrated to the unknown perturbation, thereby realizing drug function prediction, drug similarity prediction and other functions. Therefore, for drug efficacy prediction, the input data is the perturbation differential expression, and the reference data is the perturbation reference database.

[0010] In the present invention, the guiding personalized medication refers to recommending appropriate drugs or evaluating the effectiveness of drugs for different diseases. When recommending appropriate drugs, the function is based on comparing the opposite degree between the differential expression caused by the disease (input data) and the differential expression caused by the drug (perturbation reference data). If the differential expression caused by the drug can reverse the differential expression caused by the disease, then the drug can promote the disease to change to a normal state, that is, it has the ability to treat the disease, thereby realizing medication recommendation. For this function, the input data is the differential expression of the disease, and the reference data is the perturbation reference database. When evaluating the effectiveness of the drug, the function is based on the similarity between the differential expression caused by the disease (input data) and the differential expression of drug sensitivity (drug sensitivity reference data). If the differential expression caused by the disease is similar to the differential expression between drug sensitivity and insensitivity, then when in the disease state, the drug is more sensitive than in the normal state, that is, the drug is effective in the disease state and can play a role, and can be used to evaluate the effectiveness of medication. For this function, the input data is the differential expression of the disease, and the reference data is the drug sensitivity reference database.

[0011] The present invention proposes a drug recommendation method based on a multidimensional differential expression map, the drug recommendation method including a drug efficacy prediction method and / or a personalized medication guidance method. The drug recommendation method utilizes a multidimensional differential expression map reference database constructed based on the multidimensional differential expression map.

[0012] The drug recommendation method based on multi-dimensional differential expression profiles of the present invention comprises the following steps:

[0013] Step a: Obtaining a gene differential expression map of the sample to be queried: Calculating the differential expression value of the gene according to the gene expression data of the sample to be queried in different states to obtain a gene differential expression map;

[0014] Step b: Based on the gene differential expression map obtained in step a, the scores of each functional module gene set dimension used in the reference database are calculated by gene set enrichment analysis to obtain its differential expression map in different functional module dimensions; the gene differential expression map obtained in step a is integrated to obtain a multi-dimensional differential expression map of genes and functional modules of the sample to be queried;

[0015] Step c: According to the multi-dimensional differential expression map of the sample to be queried obtained in step b and the experimental purpose, a multi-dimensional differential expression map reference database is selected, and multiple similarity statistical analysis methods are used to calculate to obtain drug recommendation ranking results under different dimensions and different calculation methods;

[0016] Step d: Aggregate the multiple ranking results obtained in step c using the statistical method of model integration to ultimately obtain a reliable drug recommendation result for drug efficacy prediction and / or personalized medication guidance.

[0017] In the present invention, in the step c, the similarity calculation includes the following steps:

[0018] Step c-1: Calculate the similarity score of the input multidimensional differential expression map and the multidimensional differential expression map in the reference database using a statistical analysis method for testing similarity, including: multidimensional differential expression maps such as genes, GO functions, signaling pathways, transcriptional regulation, protein-protein interactions, etc., and perform similarity comparison analysis with the differential expression maps of the same dimension in the corresponding reference database;

[0019] The statistical analysis method of similarity refers to a method that can quantitatively characterize the similarity or correlation between data, including: Pearson correlation test, Spearman rank correlation test, Kolmogorov-Smirnov test, mutual information, cosine distance;

[0020] Step c-2: Calculate and obtain the recommended ranking results of multiple drugs under the combination conditions of different similarity methods in different dimensions according to step c-1; and / or,

[0021] In the present invention, in the step d, the ranking results are aggregated using the statistical method of model integration, which means that the ranking results of different models can be integrated so that the integrated ranking results can integrate the learning capabilities of each model and improve the generalization ability of the drug recommendation method; preferably, the statistical methods include: voting, averaging, stacking, non-cross stacking, and robust ranking integration method.

[0022] The drug recommendation method of the present invention includes a drug efficacy prediction method and / or a personalized medication guidance method.

[0023] In the drug efficacy prediction method of the present invention, a perturbation gene differential expression map is input, a perturbation reference database is selected, and a ranking list of similar drugs from the reference database is output; the smaller the ranking and the more statistically significant the drug, the stronger the similarity with the input perturbation drug efficacy;

[0024] In the personalized medication guidance method of the present invention, (a) by inputting the disease gene differential expression map, a perturbation reference database is selected, and a ranking list of drugs with opposite perturbations from the reference database is output; or, (b) by inputting the disease gene differential expression map, a drug sensitivity reference database is selected, and a ranking list of sensitive drugs from the reference database is output; the smaller the ranking and the more statistically significant the drug in the drug ranking list obtained by (a) or (b), the more potential it has for treating the input disease.

[0025] In the present invention, the method for constructing the multi-dimensional differential expression map reference database comprises the following steps:

[0026] Step I: Obtaining gene differential expression profiles;

[0027] Step II: constructing a differential expression map of the functional module gene set: based on the gene differential expression map obtained in step I, the scores in different functional module gene set dimensions are calculated by gene set enrichment analysis to obtain differential expression maps in different functional module dimensions; preferably, the different functional modules include GO functions, signaling pathways, transcriptional regulation, and protein-protein interactions;

[0028] Step III: Based on the gene differential expression map described in step I and the functional module gene set differential expression map described in step II, a multidimensional differential expression map of genes and functional modules including GO functions, signaling pathways, transcriptional regulation, and protein-protein interactions is constructed, i.e., a multidimensional differential expression map reference database.

[0029] Preferably, the gene differential expression map includes: a perturbation gene differential expression map and a drug-sensitive gene differential expression map;

[0030] Preferably, the acquisition of the perturbed gene differential expression map comprises: acquiring gene expression data before and after the perturbation from a perturbation-related database and / or literature, calculating the differential expression value of each gene before and after the perturbation, and obtaining the perturbation gene differential expression map;

[0031] Preferably, the acquisition of the drug-sensitive gene differential expression map comprises: acquiring baseline gene expression data from a database and / or literature containing drug response status or sensitivity information, calculating the differential expression value of each gene in different response states, and obtaining the drug-sensitive gene differential expression map;

[0032] Preferably, the different response states include: a drug-sensitive group and a drug-insensitive group;

[0033] Preferably, the functional module gene set differential expression map includes: a perturbation functional module gene set differential expression map and a drug sensitivity functional module gene set differential expression map;

[0034] The multidimensional differential expression map constructed corresponding to the perturbation gene differential expression map is the perturbation reference database; the multidimensional differential expression map constructed corresponding to the drug sensitivity gene differential expression map is the drug sensitivity reference database;

[0035] Preferably, the gene expression data is obtained by a gene expression analysis method; preferably, the gene expression analysis method includes: transcriptome detection and analysis, proteome detection and analysis; preferably, the transcriptome detection includes: gene chip, RNA-Seq;

[0036] Preferably, the method for obtaining the differential expression value of the gene expression data and the corresponding control expression data comprises: log FC and z-score;

[0037] Preferably, the gene expression data before and after the disturbance refers to the gene expression data of the sample treated with exogenous stimulation and the corresponding control untreated sample; preferably, the exogenous stimulation includes: compound treatment, gene knockout, gene knockdown or gene overexpression, environmental exposure; preferably, the compound includes a mixture;

[0038] Preferably, the gene set enrichment analysis includes over-representation analysis ORA, function set scoring FCS, and pathway topology analysis PT; preferably, the over-representation analysis ORA is: based on a group of genes of interest, a statistical model is used to find a preset gene function set that is significantly enriched in the group of genes; and / or, the function set scoring FCS is: based on the whole genome genes and their corresponding expression levels or differential expression level information, a statistical model is used to find a preset gene function set that is significantly enriched in the whole genome; and / or, the pathway topology analysis PT is: the position of the gene in the pathway, the connectivity with other genes and the regulatory effect information are combined to evaluate the sharing of the gene in the functional gene set and assign corresponding weights, and finally integrate it into the enrichment analysis.

[0039] Preferably, the disturbance-related database and the database of drug response status or sensitive information include: CMap, LINCS, DrugMatrix, GDSC, CTRP, NCI-60; and / or,

[0040] The functional modules refer to functional modules including GO function, signaling pathway, transcriptional regulation, and protein-protein interaction.

[0041] The present invention also proposes a method for constructing a multi-dimensional differential expression map reference database, which establishes a reference database containing multi-dimensional (gene and functional modules including GO function, signaling pathway, transcriptional regulation, protein-protein interaction, etc.) differential expression maps before and after disturbance or related to drug sensitivity based on gene differential expression maps (such as differential expression maps caused by compound perturbations, etc. or differential expression maps caused by drug sensitivity differences). The drug recommendation method is based on the reference database constructed by the method of the present invention.

[0042] The method for constructing a multi-dimensional differential expression map reference database of the present invention comprises the following steps:

[0043] Step I: Obtain gene differential expression profile:

[0044] The gene differential expression map includes: a perturbation differential gene expression map and a drug-sensitive gene differential expression map;

[0045] Wherein, the acquisition of the perturbation gene differential expression map includes: acquiring gene expression data before and after the perturbation from a perturbation-related database and / or literature, calculating the differential expression value of each gene before and after the perturbation, and obtaining the perturbation gene differential expression map; the acquisition of the drug-sensitive gene differential expression map includes: acquiring baseline gene expression data from a database and / or literature containing drug response status or sensitivity information, calculating the differential expression value of each gene in different response states, and obtaining the drug-sensitive gene differential expression map; preferably, the different response states include: a drug-sensitive group and a drug-insensitive group; and / or,

[0046] The gene expression data is obtained by a gene expression analysis method; the gene expression analysis method includes: transcriptome detection and analysis, proteome detection and analysis; the transcriptome detection includes: gene chip, RNA-Seq; and / or,

[0047] The differential expression values ​​based on the gene expression data and the corresponding control expression data can be obtained by a variety of methods, including Log fold change (log FC), z-score; and / or,

[0048] The log FC indicates the fold change of the gene expression. If the log FC is greater than 0, it indicates that the gene is up-regulated, and if it is less than 0, it indicates that the gene is down-regulated. The calculation formula is as follows:

[0049]

[0050] Among them, X a is the average expression value of the gene in the sample in state a; X b is the average expression value of the gene in samples of state b;

[0051] The calculation formula of the z-score is as follows:

[0052]

[0053] Among them, X a is the average expression value of the gene in the sample in state a; X b is the average expression value of the gene in the b-state samples, SD b is the standard deviation of the gene expression value in the b-state sample; for the perturbed gene differential expression map, the a-state is after the compound perturbation, and the b-state is before the compound perturbation. For the drug-sensitive gene differential expression map, the a-state is drug-sensitive, and the b-state is drug-insensitive.

[0054] The gene expression data before and after the disturbance refers to the gene expression data of the sample treated with exogenous stimulation and the corresponding control untreated sample; the exogenous stimulation can be compound treatment, gene knockout, gene knockdown or overexpression, environmental exposure and other conditions. The compound includes a mixture.

[0055] The drug sensitivity information refers to the ability to quantitatively measure the sensitivity of the same drug to different samples, and the samples can be grouped (sensitive vs. insensitive) according to the quantitative index.

[0056] The perturbation-related databases and databases of drug response status or sensitive information include: CMap, LINCS, DrugMatrix, GDSC, CTRP, and NCI-60.

[0057] Step II: Constructing the differential expression map of the functional module gene set: Based on the gene differential expression map obtained in step I, the scores in different functional module gene set dimensions are calculated by gene set enrichment analysis to obtain the differential expression maps in different functional module dimensions;

[0058] The functional module gene set differential expression map includes: a perturbation functional module gene set differential expression map and a drug sensitivity functional module gene set differential expression map;

[0059] The functional modules refer to GO functions, signaling pathways, transcriptional regulation, protein-protein interactions and other functional modules;

[0060] The functional module gene sets include transcriptional factor target gene sets, GO biological process, GO cellular component, GO molecular function, PID pathway interactions, STRING protein-protein interaction, KEGG pathway, IPA pathway, Reactome biological pathway, and ImmuneSigDB immunologic signature gene sets.

[0061] The gene set enrichment analysis includes over-representation analysis ORA, function set scoring FCS and pathway topology analysis PT: wherein the over-representation analysis ORA is: based on a group of genes of interest, a statistical model is used to find a preset gene function set that is significantly enriched in the group of genes; the function set scoring FCS is: based on the whole genome genes and their corresponding expression levels or differential expression level information, a statistical model is used to find a preset gene function set that is significantly enriched in the whole genome; the pathway topology analysis PT is: the position of the gene in the pathway, the connectivity with other genes and the regulatory effect information are combined to evaluate the sharing of the gene in the functional gene set and assign corresponding weights, and finally integrate them into the enrichment analysis.

[0062] Step III: Based on the gene differential expression map described in step I and the functional module gene set differential expression map described in step II, a multidimensional differential expression map of genes and functional modules including GO functions, signaling pathways, transcriptional regulation, protein-protein interactions, etc. is constructed, i.e., a multidimensional differential expression map reference database.

[0063] Corresponding to the perturbed gene differential expression map in step I, the constructed multi-dimensional differential expression map is a perturbation reference database (such as CMap, LINCS, etc.) that can be used for drug efficacy prediction or guiding personalized medication.

[0064] Corresponding to the drug-sensitive gene differential expression map in step I, the constructed multi-dimensional differential expression map is a drug sensitivity reference database (such as GDSC, NCI-60, CTRP, DrugMatrix, etc.), which can be used to guide personalized medication.

[0065] The present invention also proposes the application of the above-mentioned multidimensional differential expression map reference database and its construction method, the drug recommendation method, and the drug recommendation system in predicting drug efficacy and guiding personalized medication. The present invention is based on a method for predicting drug efficacy and guiding personalized medication based on a multidimensional differential expression map, such as drug efficacy prediction including but not limited to predicting similar medication, drug reuse and target prediction, or the effectiveness of drug treatment of tumors; it selects a perturbation reference database. For example, the personalized medication guidance refers to recommending appropriate drugs for specific diseases, such as different tumors; it selects a perturbation reference database or a drug sensitivity reference database.

[0066] The present invention also proposes a drug recommendation system for drug efficacy prediction and / or personalized medication guidance based on a multi-dimensional differential expression map, the system comprising: a reference database module, a gene differential expression map module, a multi-dimensional differential expression map module, a drug recommendation ranking module and a drug recommendation result module; preferably,

[0067] The reference database module is used to construct a multi-dimensional differential expression map reference database;

[0068] The gene differential expression map module is used to obtain gene expression data, and calculate the gene differential expression value under the data to obtain the gene differential expression map;

[0069] The multi-dimensional differential expression map module is used to calculate the differential expression map under the functional module gene set, and combine the gene differential expression map to obtain a multi-dimensional differential expression map;

[0070] The drug recommendation ranking module is used to calculate the drug recommendation ranking results under different dimensions and different calculation methods using a similarity statistical analysis method;

[0071] The drug recommendation result module is used to obtain reliable drug recommendation results using model-integrated statistics for drug efficacy prediction and / or personalized medication guidance.

[0072] Furthermore, the system also includes: a homepage module and a help module; wherein,

[0073] The homepage module is used to introduce the system;

[0074] The help module is used to introduce how to use the system.

[0075] In a specific implementation scheme, in the drug efficacy prediction, a perturbation gene differential expression map is input, a perturbation reference database is selected, and a ranking list of similar drugs derived from the reference database is output; the smaller the ranking and the more statistically significant the drug, the stronger the similarity with the input perturbation drug efficacy.

[0076] In a specific embodiment, in the personalized medication guidance method, (a) by inputting the disease gene differential expression map, a perturbation reference database is selected, and a ranking list of drugs with opposite perturbations from the reference database is output, or, (b) by inputting the disease gene differential expression map, a drug sensitivity reference database is selected, and a ranking list of sensitive drugs from the reference database is output; the smaller the ranking of the drug ranking obtained by (a) or (b) and the more statistically significant the drug, the greater the potential for treating the input disease.

[0077] The present invention also proposes a computer device, comprising an input device, a processor, a memory and an output device, wherein the input device, the processor, the memory and the output device are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the above method of the present invention.

[0078] The present invention further proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and is characterized in that the computer program implements the above method of the present invention when executed by a processor.

[0079] The beneficial effects of the present invention include: for the first time, the present invention creatively infers the association between genes and functional information including GO functions, signal pathways, transcriptional regulation, protein-protein interactions, etc., and finally achieves more accurate drug efficacy prediction and personalized medication guidance compared with similar work, benefiting from integrated statistical methods. The present invention provides interactive web pages, and the computational complexity is controllable, which is convenient for users to perform convenient and friendly calculations and queries. Moreover, the present invention is not limited to a certain type of cancer, and does not require any disease-related information. Only through differential expression maps, according to the purpose of the experiment, drug recommendations can be made, and the applicability is wide. The present invention has good scalability. With the accumulation and expansion of reference databases, the methodological framework can be rapidly expanded to the research and application of more complex systemic diseases beyond tumors, including for drug efficacy prediction and personalized medication guidance. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 This is a flow chart of a method for predicting drug efficacy and / or guiding personalized medication based on a multi-dimensional differential expression profile in a specific embodiment of the present invention.

[0081] Figure 2 It is a schematic diagram of predicting drug efficacy and / or guiding personalized medication based on multi-dimensional differential expression maps in a specific embodiment of the present invention.

[0082] Figure 3 This is a diagram of the operating results of a drug recommendation system for predicting drug efficacy and / or guiding personalized medication according to the present invention.

[0083] Figure 4 This is the ROC curve of the results of recommending similar drugs of histone acetylase inhibitors by different methods in Example 2 of the present invention.

[0084] Figure 5This is a survival analysis chart of ovarian cancer patients in Example 3 of the present invention. The horizontal axis is the survival time, and the vertical axis is the survival rate (i.e., the proportion of samples that are still alive in the current group of patients); among all patients, patients who received reasonable drug treatment (using the first 20 drugs in the recommended list of the present invention) were divided into the correct treatment group, and other patients were divided into the wrong treatment group. The truncated data refers to the samples that have not died at the current time of tracking; the risk ratio refers to the death risk ratio of the two groups of patients. The 0.480 in this figure means that the death risk of patients in the correct treatment group is 0.480 times that of the wrong treatment group; p-value<0.05 indicates the statistical significance of the difference in the death risk of the two groups of patients (Log-rank test). DETAILED DESCRIPTION

[0085] The invention is further described in detail with reference to the following specific embodiments and drawings. The process, conditions, experimental methods, etc. for implementing the present invention, except for the contents specifically mentioned below, are all common knowledge and common common sense in the art and are not particularly limited by the present invention.

[0086] Example 1: Preparing a multi-dimensional differential expression profile reference database

[0087] 1.1 Obtain expression map data from public databases. Depending on the experimental scheme, one type is the expression map before and after drug perturbation, and calculate the differential expression value of the gene before and after perturbation; the other type is the baseline gene expression map data, which can be divided into two groups of samples with different responses to the same drug based on drug sensitivity, and calculate the differential expression value of the gene between the sensitive group and the insensitive group. The first category is defined as the differential expression map of perturbed genes, and the second category is the differential expression map of drug-sensitive genes, collectively referred to as gene differential expression maps.

[0088] 1.1.1 Obtaining differential expression profiles of perturbed genes:

[0089] (1) Obtain the expression profiles of cell lines before and after drug perturbation from the public database CMap. According to the database design, the gene expression data generated by the CMap database is generated after a compound is treated with a cell line at a certain concentration for a specific period of time. Since the treatment time of most samples is the same and many treatment doses are only one, the differences in dose and treatment time are not considered for the CMap data. The data of the same compound treating the same cell line are merged to obtain compound-cell line grouping. The expression profile data of the baseline (control sample is compound DMSO treatment) is compared, and the differential expression value log FC of the sample after drug treatment and the baseline sample is calculated using the limma package to obtain the differential expression profile of the compound-cell line pair.

[0090] (2) Download the differential expression maps of cell lines before and after compound perturbation from LINCS - This database provides the differential expression values ​​of samples after compound perturbation treatment and baseline samples after Z-score normalization. Since the number of compounds used in LINCS and the corresponding cell lines, treatment time, and treatment dose combinations are too complex, only cancer drug treatment data are selected, and the differential expression maps corresponding to the samples after cancer drug treatment and baseline are finally obtained from LINCS.

[0091] 1.1.2 Obtaining differential expression profiles of drug-sensitive genes:

[0092] The public database GDSC was selected to construct a differential expression map of drug-sensitive genes. The GDSC database includes two sets of data sets, GDSC1 and GDSC2, and provides the IC50 values ​​of the cell lines involved in the above two sets of data sets, which can be used to measure the sensitivity to drugs. Download the baseline expression map data of each cell line and the IC50 value under the corresponding drug.

[0093] For each drug of GDSC1 and GDSC2, the IC50 values ​​in different cell lines were sorted from small to large, and the top 25% and bottom 25% of the cell lines were identified as drug-sensitive cell lines and insensitive cell lines, respectively. Based on this grouping, the differential expression values ​​of the drug-sensitive cell lines and the insensitive cell lines were calculated using the limma package, and finally the differential expression matrix of the drug-sensitive and insensitive groups in GDS1 and GDSC2 was obtained.

[0094] 1.2 Obtain functional gene set information from the public functional gene set database, and use gene set enrichment analysis to calculate the scores of different functional module gene sets under different states in step 1.1.

[0095] Considering that changes in signal pathways and transcriptional regulatory networks can better characterize the differences in the functional characteristics of organisms under different conditions, the present invention reduces the dimension of the differential expression map at the gene level and maps it to the functional gene set level. According to the perturbation and drug sensitivity differential expression maps at the gene level obtained above, the GSEA function in the R package clusterProfiler is used to calculate the scores under the functional module gene sets such as KEGG, MSigDB-TFT (transcription factor targtes), GO-MF, and GO-CC.

[0096] 1.3 Integrate the gene differential expression values ​​and functional gene set scores under different states in 1.1 and 1.2, and construct a multidimensional differential expression map as a reference differential expression map. Corresponding to the first type of expression map data in step 1.1, the constructed multidimensional differential expression map is a perturbation reference differential expression map (CMap, LINCS), which can be used to predict drug efficacy or guide personalized medication; corresponding to the second type of drug sensitive gene differential expression map in 1.2, the constructed multidimensional differential expression map is a drug sensitive reference differential expression map (GDSC), which can be used to guide personalized medication.

[0097] Figure 1 is a flow chart of the method and application of the present invention, Figure 2 The figure is a schematic diagram of the method and application of the present invention. The method of the present invention is based on the differential expression map of genes (such as differential expression maps caused by disease states or drug disturbances, etc.), and calculates the multi-dimensional differential expression map of gene sets in functional modules such as pathways, transcriptional regulation, and GO pathways, and constructs a multi-dimensional differential expression map to be queried, and then compares it with the perturbation reference database or drug sensitivity reference database established by the present invention. After obtaining the recommendation results of multiple dimensions by the similarity measurement method, the statistical method of model integration is aggregated to obtain a comprehensive drug recommendation ranking result, which can be used to predict drug efficacy (including drug similarity prediction, drug function prediction, etc.) and guide personalized medication.

[0098] S1, obtaining the gene expression data of samples before and after a certain state (disease or drug disturbance, etc.) and calculating the gene differential expression value under the data or directly obtaining the gene differential expression value.

[0099] S2, based on the gene differential expression values ​​obtained in S1, the gene set enrichment analysis is used to calculate the scores of the functional module gene sets in different dimensions. The functional module gene sets are the same as the functional module gene sets used in the reference differential expression map in step 1.2, and the multi-dimensional differential expression map of the state (disease or drug disturbance, etc.) is obtained;

[0100] S3, select the reference differential expression profile according to the experimental purpose.

[0101] If the purpose of the experiment is to predict drug efficacy (including but not limited to drug similarity prediction, drug function prediction), etc., then the perturbation reference differential expression map may be selected.

[0102] If the purpose of the experiment is to guide personalized medication, you can choose a perturbation reference differential expression map or a drug sensitivity reference differential expression map.

[0103] Figure 1 The process shown is a method of the present invention for predicting drug efficacy and / or guiding personalized medication based on a multi-dimensional differential expression map, and the input only needs to be a differential gene expression map.

[0104] S4, based on the multi-dimensional differential expression map of the state (disease or disturbance, etc.) obtained in S2 and the reference differential expression map, a similarity statistical analysis method is used to calculate and obtain recommendation results under different dimensions and different calculation methods;

[0105] The similarity statistical analysis method includes the following steps:

[0106] (1) Calculate the similarity score between the input multi-dimensional differential expression map and the constructed reference differential expression map using a statistical analysis method for testing similarity, including five differential expression maps: gene, pathway (KEGG), cell function (GO-BP), cell location (GO-CC), and transcriptional regulatory target (TFT), and perform similarity comparison analysis with the expression maps of the same dimension in the corresponding reference differential expression map;

[0107] The similarity statistical analysis method described in S4 refers to a method that can quantitatively characterize the degree of similarity or correlation between data. In this method, the Spearman rank correlation test and the Kolmogorov-Smirnov (KS) test are used.

[0108] The formula for the Spearman correlation test is as follows:

[0109]

[0110] Among them, d i is the position difference between the corresponding identical elements (such as genes, functional gene sets) in the input multidimensional differential expression map and the reference differential expression map after sorting, n is the total number of all observed elements (total number of genes, total number of functional gene sets), and ρ is the Spearman rank correlation coefficient.

[0111] The KS test formula is as follows:

[0112]

[0113]

[0114] First, the genes or functional gene sets in the input data are sorted, and the largest or smallest m data corresponding to the reference differential expression map are selected to obtain the intersection of set A, which has Num_a genes in total, and the total number of genes or functional gene sets in the corresponding reference differential expression map is Num_b; set A is sorted according to the input data value, i represents the rank after sorting, i_position is the ranking of the corresponding gene or functional gene set in the reference database, and a and b are calculated respectively according to the above two formulas.

[0115] Remember KS topKS is the value calculated from the largest m data corresponding to the reference data. bottom The value calculated for the smallest m data corresponding to the reference data.

[0116] If a is greater than b, then KS top or KS bottom The value is a if it is not, otherwise it is b.

[0117] If KS top and KS bottom The product of is less than 0, then KS=KS top -KS bottom ; otherwise KS=0

[0118] For the reference differential expression map under each condition in the reference data, the corresponding Spearman rank correlation coefficient or KS value can be obtained, and the sorting can be performed according to the value.

[0119] (2) Multiple drug recommendation results are calculated under the combination of different similarity methods in different dimensions;

[0120] S5, uses the statistical method of model integration to aggregate the multiple ranking results of S4, and finally obtains a comprehensive drug recommendation ranking result for drug efficacy prediction or personalized medication guidance recommendation.

[0121] The statistical method of model integration described in S5 is calculated using the RRA method in this scheme.

[0122] For the recommended drugs, the present invention uses ATC (Anatomical Therapeutic Chemical) - the anatomical, therapeutic and chemical classification of drugs to annotate drug information and drug-target, drug-disease information from drug target databases such as Drugbank, TTD (Therapeutic target database), DrugRepurposing and Medical Rubik's Cube to annotate the drugs, so as to evaluate the reliability of the recommendation system.

[0123] The following specific examples respectively specifically illustrate the drug efficacy prediction (including drug similarity prediction, drug function prediction) and the guidance of personalized medication (drug effectiveness prediction) based on the above-mentioned method.

[0124] Example 2: Pharmacodynamic prediction (drug similarity prediction) - Verification of the rationality of the method of the present invention in recommending histone deacetylase inhibitors (HDAC inhibitors) (pharmacological prediction - recommendation of drugs with similar mechanisms)

[0125] 1. Obtain the differential expression map of HDAC inhibitor perturbation in CMap

[0126] Drugs with HDAC inhibitors as their mechanism of action were retrieved from various drug target databases (Drugbank, TTD, DrugRepurposing, and Pharmaceutical Cube, etc.). Based on these drug lists, the perturbation gene differential expression data of HDAC inhibitors in different cell lines were screened from CMap. Subsequently, any perturbation gene differential expression data was used to query the present invention to explore whether the remaining perturbations were recommended first.

[0127] 2. Recommending Similar Drugs Using the Method of the Present Invention and Similar Methods

[0128] The differential expression of genes perturbed by HDAC inhibitors in step 1 is used as input respectively. Since the experimental purpose of the present invention is to predict drug efficacy (find similar drugs), the present invention selects CMap as a reference to obtain similar drug recommendation results under different HDAC inhibitor perturbations. The recommendation results are given in the form of p-value sorting from small to large.

[0129] The specific steps are as follows:

[0130] 1. Obtain the expression values ​​before and after HDAC inhibitor perturbation from the CMap database and calculate the differential gene expression values ​​based on them;

[0131] 2. Based on the gene differential expression values ​​obtained in step 1, gene set enrichment analysis (GSEA function in R package clusterProfiler) was used to calculate the scores of functional gene sets of different dimensions (KEGG, MSigDB-TFT (transcription factor targtes), GO-MF, GO-CC) to obtain the multi-dimensional differential expression map under the above perturbations;

[0132] 3. Select CMap as the reference database.

[0133] 4. Based on the multidimensional differential expression map obtained in step 2 and the CMap reference database, the Spearman correlation test method and the KS test method were used to calculate the similarity.

[0134] The formula for the Spearman correlation test is as follows:

[0135]

[0136] Among them, d iis the position difference between the multidimensional differential expression map after the input HDAC inhibitor perturbation and the corresponding same elements (such as genes, functional gene sets) in the CMap reference database after sorting, n is the total number of all elements observed (total number of genes, total number of functional gene sets), and ρ is the Spearman rank correlation coefficient.

[0137] The KS test formula is as follows:

[0138]

[0139]

[0140] First, sort the genes or functional gene sets in the input data, select the largest or smallest m and take the intersection with the corresponding data in the CMap reference database to get set A, which has Num_a genes in total, and the total number of genes or functional gene sets in the corresponding reference database is Num_b; sort set A according to the input data value, i represents the rank after sorting, i_position is the ranking of the corresponding gene or functional gene set in the reference database, and calculate a and b respectively according to the above two formulas.

[0141] Remember KS top KS is the value calculated from the largest m data corresponding to the reference data. bottom The value calculated for the smallest m data corresponding to the reference data.

[0142] If a is greater than b, then KS top or KS bottom The value is a if it is not, otherwise it is b.

[0143] If KS top and KS bottom The product of is less than 0, then KS=KS top -KS bottom ; otherwise KS=0

[0144] For each reference spectrum under perturbation in the reference database, the corresponding Spearman rank correlation coefficient or KS value can be obtained, and the ranking can be performed according to the value.

[0145] 5. The RRA method was used to integrate the recommendation ranking results of the two methods under five types of data (genes, four functional gene sets) to obtain a complete drug recommendation result.

[0146] At the same time, using the above-mentioned perturbation differential expression map as input, the present invention uses two sets of tools, SignatureSearch [DuanY, Evans DS, Miller RA, Schork NJ, Cummings SR, Girke T. SignatureSearch: environment for gene expression signature searching and functional interpretation. Nucleic Acids Res. 2020.] and Dr Insight [Chan J, Wang X, Turner JA, Baldwin NE, Gu J. Breaking the paradigm: Dr Insight empowers signature-free, enhanced drug repurposing. Bioinformatics. 2019.] to recommend similar drugs respectively, and the recommended results are also given in the form of p-value sorting from small to large. Figure 3 This is a graph showing the operation results of the drug recommendation system of the present invention after a certain HDAC inhibitor perturbation gene differential expression map is input.

[0147] 3. Recommended Performance Evaluation

[0148] In this embodiment, the perturbation expression map of drugs whose mechanism of action is HDAC inhibitor is used as input. According to the principle of the present invention, the perturbation of drugs that are also HDAC inhibitors should be recommended at the top of the results.

[0149] The recommended results were combined and sorted from small to large according to the recommended p-value; all drugs with the mechanism of action of HDAC inhibitors were marked as 1 (drugs with the same mechanism of action were considered to be correctly recommended), and the rest were marked as 0, and then the model AUC (Area Under the Curve) was calculated. The ROC curve was drawn based on the result, and the area under the curve AUC of the present invention was 0.9721, which was higher than the results of the other two methods (the AUC of signatureSearch was 0.9215, the AUC of Drinsight was 0.9569, and the ROC curve obtained in Example 1 was shown in FIG. Figure 4 , indicating that the HDAC inhibitor-like drugs recommended by the present invention are more effective.

[0150] Example 3: Personalized medication - Verifying the rationality of the breast cancer treatment drug recommendation by the method of the present invention (personalized medication - recommended medication)

[0151] 1. Collecting gene expression data from breast cancer samples

[0152] Gene expression data of breast cancer and control samples were downloaded from GEO (https: / / www.ncbi.nlm.nih.gov / geo / query / acc.cgi) (accession: GSE42568, [HG-U133_Plus_2] Affymetrix Human Genome U133 Plus 2.0 Array), and gene expression data of 104 breast cancer samples and 17 normal breast tissue samples were obtained.

[0153] 2. Calculate the multi-dimensional differential expression profile of breast cancer samples

[0154] Based on the breast cancer samples and control samples collected in step 1, the fold-change of each gene in 104 breast cancer samples relative to 17 normal breast samples was calculated to obtain the gene differential expression map under this state.

[0155] Based on the gene differential expression map, the gene set enrichment analysis of the present invention is used to calculate the scores in different functional gene sets (KEGG, GO-BP, GO-CC, TFT) to obtain differential expression maps of gene sets in different functional module dimensions; the gene differential expression maps are integrated to obtain a multi-dimensional differential expression map of breast cancer.

[0156] 3. Recommending therapeutic drugs using the method of the present invention

[0157] The multi-dimensional differential expression map of breast cancer in step 2 is used as input. Since the purpose of the experiment is to guide personalized medication, the present invention selects LINCS, which is a full database of cancer drug perturbations, as a reference database to obtain the recommended results of therapeutic drugs for breast cancer. The recommended results are given in the form of p-value sorting from small to large.

[0158] The specific steps are as follows:

[0159] 1. Based on the multidimensional differential expression map obtained in step 2 and the LINCS reference data, the Spearman correlation test method and the KS test method were used to calculate the similarity.

[0160] The formula for the Spearman correlation test is as follows:

[0161]

[0162] Among them, d i is the position difference between the same elements (such as genes, functional gene sets) in the input breast cancer multidimensional differential expression map and the LINCS reference differential expression map after sorting, n is the total number of all elements observed (total number of genes, total number of functional gene sets), and ρ is the Spearman rank correlation coefficient.

[0163] The KS test formula is as follows:

[0164]

[0165]

[0166] First, sort the genes or functional gene sets in the input data, select the largest or smallest m data corresponding to the LINCS reference database and take the intersection to get set A, which has Num_a genes in total, and the total number of genes or functional gene sets in the corresponding reference database is Num_b; sort set A according to the input data value, i represents the rank after sorting, i_position is the ranking of the corresponding gene or functional gene set in the reference database, and calculate a and b respectively according to the above two formulas.

[0167] Remember KS top KS is the value calculated from the largest m data corresponding to the reference data. bottom The value calculated for the smallest m data corresponding to the reference data.

[0168] If a is greater than b, then KS top or KS bottom The value is a if it is not, otherwise it is b.

[0169] If KS top and KS bottom The product of is less than 0, then KS=KS top -KS bottom ; otherwise KS=0

[0170] For each reference spectrum under perturbation in the reference database, the corresponding Spearman rank correlation coefficient or KS value can be obtained, and the ranking can be performed according to the value.

[0171] 2. The RRA method was used to integrate the recommendation ranking results of the two methods under five types of data (genes, four functional gene sets) to obtain a complete drug recommendation result.

[0172] IV. Recommended Performance Evaluation

[0173] The top 10 recommendations were selected to find evidence for their use in treating breast cancer. The top 10 included ponatinib (1, 6), crizotinib (2), dasatinib (3), palbociclib (4, 5), nilotinib (7), vandetanib (8), mitoxantrone (9), and idarubicin (10), with rankings in brackets. A total of 8 drugs were involved.Among the top 10 recommended drugs, 4 (crizotinib, dasatinib, palbociclib and vandetanib) have indications including breast cancer (data from drug target databases such as Drugbank, TTD (Therapeutic target database), DrugRepurposing and Pharmaceutical Cube), and the remaining 4 drugs (ponatinib, nilotinib, mitoxantrone and idarubicin) have been reported in the literature to have the potential to be breast cancer drugs: ponatinib can be used as a new drug to inhibit breast cancer lung metastasis [Shao W, Li S, Li L, et al. Chemical genomics reveals inhibition of breast cancer lungmetastasis by Ponatinib via c-Jun. Protein Cell. 2019.]; nilotinib can induce apoptosis and inhibit migration of breast cancer cells [Wang S, Xie Y, Bao A, et al. Nilotinib, a Discoidin domain receptor 1 (DDR1) inhibitor, induces apoptosis and inhibits migration in breast cancer. Neoplasma. 2021.】;mitoxantrone can be combined with vinorelbine to treat metastatic breast cancer

VogelCL. Combination chemotherapy with vinorelbine (Navelbine) and mitoxantrone formetastatic breast cancer: a review. Semin Oncol. 1995.

Bastholt L, Dalmark M, Jakobsen A, Gadeberg CC, Sandberg E, Mouridsen HT. Oral idarubicin in the treatment of advanced breast cancer. Acta Oncol. 1989.

[0174] Table 1 Top 10 recommended drugs for breast cancer patients

[0175]

[0176] Note:

[0177] 1) Only fill in the clinical stage for drugs for the treatment of breast cancer; if it is not a drug for the treatment of breast cancer, fill in the relevant literature;

[0178] 2) “ / ” means no need to fill in

[0179] Example 4: Prediction of drug effectiveness - Prediction of drug effectiveness for ovarian cancer patients using the method of the present invention

[0180] 1. Collecting gene expression data and clinical data from ovarian cancer patient samples

[0181] Gene expression data (log2(TPM+0.001)) of ovarian cancer (OV) patients and control samples, prognosis data of patients (including survival time, survival status, etc.) and medication information of patients were downloaded from the data resources provided by UCSC (https: / / xena.ucsc.edu / public / ). The gene expression data and clinical data of patients were derived from TCGA (The Cancer), and the gene expression data of control samples were derived from GTEx (Genotype-Tissue Expression Project). Finally, gene expression data of 274 ovarian cancer samples and 88 normal ovarian tissue samples were obtained, as well as medication information (1396 groups) and prognosis data of corresponding patient samples.

[0182] 2. Use public databases to unify drug names

[0183] From the drug database DrugBank (https: / / go.drugbank.com / ), search for the official names of the drugs used by the patient in DrugBank and the official names of the drugs used in the reference data in DrugBank.

[0184] 3. Calculate the differentially expressed gene list of patients relative to normal samples

[0185] For each cancer sample, the fold-change value of each gene is calculated using its own gene expression data and the gene expression data of the control sample (in this embodiment, the mean value of the gene expression data of 88 normal samples is used as the control data of each cancer sample. In fact, the best choice should be the gene expression data of the corresponding adjacent cancer tissue). For each cancer sample, the fold-change values ​​of all genes can be calculated, and this list is used as the input for the next step of drug recommendation.

[0186] 4. Prediction of drug activity for each cancer sample

[0187] Using the differential expression map of each cancer sample in step three as input, the gene set enrichment analysis of the present invention is used to calculate the scores in different functional gene sets (KEGG, GO-BP, GO-CC, TFT), and the differential expression maps of different pathways and transcriptional regulation dimensions are obtained; the gene differential expression maps are integrated to obtain a multi-dimensional differential expression map of each cancer sample. GDSC is selected as the reference database to obtain the drug sensitivity prediction results under different cancer samples. The prediction results are given in the form of p-value sorting from small to large. The specific steps are as follows:

[0188] 1. Based on the multidimensional differential expression map and GDSC reference data obtained in step 4, the Spearman correlation test method and KS test method were used to calculate the similarity.

[0189] The formula for the Spearman correlation test is as follows:

[0190]

[0191] Among them, d i is the position difference between the multidimensional differential expression map of the input ovarian cancer samples and the corresponding identical elements (such as genes, functional gene sets) in the GDSC reference database after sorting, n is the total number of all elements observed (total number of genes, total number of functional gene sets), and ρ is the Spearman rank correlation coefficient.

[0192] The KS test formula is as follows:

[0193]

[0194]

[0195] First, sort the genes or functional gene sets in the input data, select the largest or smallest m and take the intersection with the corresponding data in the reference data to get set A, which has Num_a genes in total, and the total number of genes or functional gene sets in the corresponding reference database is Num_b; sort set A according to the input data value, i represents the rank after sorting, i_position is the ranking of the corresponding gene or functional gene set in the reference database, and calculate a and b respectively according to the above two formulas.

[0196] Remember KS top KS is the value calculated from the largest m data corresponding to the reference data. bottom The value calculated for the smallest m data corresponding to the reference data.

[0197] If a is greater than b, then KS top or KS bottom The value is a if it is not, otherwise it is b.

[0198] If KS top and KS bottom The product of is less than 0, then KS=KS top -KS bottom ; otherwise KS=0

[0199] For each reference spectrum under each condition in the reference data, the corresponding Spearman rank correlation coefficient or KS value can be obtained, and the ranking can be performed according to the value.

[0200] 2. The RRA method was used to integrate the recommendation ranking results of the two methods under five types of data (genes, four functional gene sets) to obtain a complete drug recommendation result.

[0201] V. Prediction Performance Evaluation

[0202] For the 274 ovarian cancer patient samples in this embodiment, if at least one of the drugs used is in the top 20 drugs in the prediction results, it is considered that the sample has received the correct treatment and is divided into the group receiving the correct treatment, otherwise it is divided into the group receiving the wrong treatment. Afterwards, the two groups of samples are subjected to survival analysis based on the prognostic information of the two groups of samples to check whether the group receiving the correct treatment has a longer survival time than the group receiving the wrong treatment. If so, it proves that the method of the present invention is effective.

[0203] According to the predicted results, among the 274 ovarian cancer patient samples, 230 samples belonged to the group receiving the correct treatment, and 44 samples belonged to the group receiving the wrong treatment. Survival analysis of these two groups showed that the p value of the log-rank test was 0.00027, and the hazard ratio (Hazard Ratio, the group receiving the correct treatment compared to the group receiving the wrong treatment) was 0.480 (95% confidence interval was 0.321 to 0.718). The survival curve of Example 4 is attached. Figure 5 According to the above results, it can be shown that the prediction method of the present invention is effective.

[0204] The effectiveness of the method was further verified by a permutation test. For 274 samples, 230 samples were randomly selected as the group receiving the correct treatment, and the remaining 44 samples were randomly selected as the group receiving the wrong treatment. The risk ratio between the two randomly divided groups and the p-value of the log-rank test were calculated; the process was repeated 1000 times, and it was found that only 2 results were better than the predicted grouping (risk ratio less than 1 and p-value less than 0.00027), and the p-value of the permutation test was 0.002. This further illustrates the effectiveness of the method of the present invention.

[0205] The protection content of the present invention is not limited to the above embodiments. Without departing from the spirit and scope of the inventive concept, changes and advantages that can be thought of by those skilled in the art are included in the present invention and are protected by the attached claims.

Claims

1. A drug recommendation method for drug efficacy prediction and / or personalized medication guidance, characterized in that: The method is based on a multi-dimensional differential expression map reference database. It infers the association with the drug mechanism or disease phenotype through the multi-dimensional information of genes and functional modules including GO function, signaling pathway, transcriptional regulation, and protein-protein interaction, and then uses the statistical method of model integration to achieve drug recommendation.

2. The drug recommendation method according to claim 1, characterized in that: The multi-dimensional differential expression map reference database includes a perturbation reference database and / or a drug sensitivity reference database; Preferably, in the drug efficacy prediction, a perturbation gene differential expression map is input, a perturbation reference database is selected, and a ranking list of similar drugs from the reference database is output; the smaller the ranking and the more statistically significant the drug, the stronger the similarity with the input perturbation drug efficacy; Preferably, in the personalized medication guidance, (a) by inputting the disease gene differential expression map, a perturbation reference database is selected, and a ranking list of drugs with opposite perturbations from the reference database is output; or, (b) by inputting the disease gene differential expression map, a drug sensitivity reference database is selected, and a ranking list of sensitive drugs from the reference database is output; preferably, the drug with a lower ranking and a statistically significant drug in the drug ranking list obtained by (a) or (b) has a greater potential for treating the input disease.

3. The drug recommendation method according to claim 1, characterized in that: The method comprises the following steps: Step a: Obtaining a gene differential expression map of the sample to be queried: Calculating the differential expression value of the gene according to the gene expression data of the sample to be queried in different states to obtain a gene differential expression map; Step b: Based on the gene differential expression map obtained in step a, the scores of each functional module gene set dimension used in the reference database are calculated by gene set enrichment analysis to obtain its differential expression map in different functional module dimensions; the gene differential expression map obtained in step a is integrated to obtain a multi-dimensional differential expression map of genes and functional modules of the sample to be queried; Step c: According to the multi-dimensional differential expression map of the sample to be queried obtained in step b and the experimental purpose, a multi-dimensional differential expression map reference database is selected, and multiple similarity statistical analysis methods are used to calculate to obtain drug recommendation ranking results under different dimensions and different calculation methods; Step d: Aggregate the multiple ranking results obtained in step c using the statistical method of model integration to ultimately obtain a reliable drug recommendation result for drug efficacy prediction and / or personalized medication guidance.

4. The drug recommendation method according to claim 3, characterized in that: In the step c, the similarity statistical analysis method comprises the following steps: Step c-1: using a statistical analysis method for testing similarity to calculate the similarity score between the input multidimensional differential expression map and the multidimensional differential expression map in the reference database, including: gene, GO function, signal pathway, transcriptional regulation, protein-protein interaction differential expression map, respectively, and the corresponding reference database in the same dimension of the differential expression map for similarity comparison analysis; preferably, the similarity statistical analysis method refers to a method that can quantitatively characterize the similarity or correlation between data, including Pearson correlation test, Spearman rank correlation test, Kolmogorov-Smirnov test, mutual information, cosine distance; Step c-2: Calculate and obtain multiple drug recommendation ranking results based on the combination of different dimensions and different similarity statistical analysis methods in step c-1; and / or, In the step d, the ranking results are aggregated using the statistical method of model integration, which means that the ranking results of different models are integrated so that the integrated ranking results can integrate the learning capabilities of each model and improve the generalization ability of the drug recommendation method; preferably, the statistical methods include voting, averaging, stacking, non-cross stacking, and robust ranking integration methods.

5. The drug recommendation method according to claim 3, characterized in that: The different states include: different disease states, states before and after disturbance, different genetic backgrounds or different drug effects; wherein, The different disease states include: health and disease; The states before and after the disturbance are before and after the exogenous stimulus treatment; The different genetic backgrounds include: the presence or absence of disease-related gene mutations; The different drug effects include: effective and ineffective, sensitive and resistant.

6. The drug recommendation method according to claim 1, characterized in that: The method for constructing the multi-dimensional differential expression map reference database comprises the following steps: Step I: Obtaining gene differential expression profiles; Step II: constructing a differential expression map of the functional module gene set: based on the gene differential expression map obtained in step I, the scores in different functional module gene set dimensions are calculated by gene set enrichment analysis to obtain differential expression maps in different functional module dimensions; preferably, the different functional modules include GO functions, signaling pathways, transcriptional regulation, and protein-protein interactions; Step III: Based on the gene differential expression map described in step I and the functional module gene set differential expression map described in step II, a multidimensional differential expression map of genes and functional modules including GO functions, signaling pathways, transcriptional regulation, and protein-protein interactions is constructed, i.e., a multidimensional differential expression map reference database.

7. The drug recommendation method according to claim 6, characterized in that: The gene differential expression map includes: a perturbation gene differential expression map and a drug sensitive gene differential expression map; Preferably, the acquisition of the perturbed gene differential expression map comprises: acquiring gene expression data before and after the perturbation from a perturbation-related database and / or literature, calculating the differential expression value of each gene before and after the perturbation, and obtaining the perturbation gene differential expression map; Preferably, the acquisition of the drug-sensitive gene differential expression map comprises: acquiring baseline gene expression data from a database and / or literature containing drug response status or sensitivity information, calculating the differential expression value of each gene in different response states, and obtaining the drug-sensitive gene differential expression map; Preferably, the different response states include: a drug-sensitive group and a drug-insensitive group; Preferably, the functional module gene set differential expression map includes: a perturbation functional module gene set differential expression map and a drug sensitivity functional module gene set differential expression map; The multidimensional differential expression map constructed corresponding to the perturbation gene differential expression map is the perturbation reference database; the multidimensional differential expression map constructed corresponding to the drug sensitivity gene differential expression map is the drug sensitivity reference database; Preferably, the gene expression data is obtained by a gene expression analysis method; preferably, the gene expression analysis method includes: transcriptome detection and analysis, proteome detection and analysis; preferably, the transcriptome detection includes: gene chip, RNA-Seq; Preferably, the method for obtaining the differential expression value of the gene expression data and the corresponding control expression data comprises: logFC and z-score; Preferably, the gene expression data before and after the disturbance refers to the gene expression data of the sample treated with exogenous stimulation and the corresponding control untreated sample; preferably, the exogenous stimulation includes: compound treatment, gene knockout, gene knockdown or gene overexpression, environmental exposure; preferably, the compound includes a mixture; and / or, The gene set enrichment analysis includes over-representation analysis ORA, function set scoring FCS, and pathway topology analysis PT; preferably, the over-representation analysis ORA is: based on a group of genes of interest, a preset gene function set that is significantly enriched in the group of genes is found using a statistical model; and / or, the function set scoring FCS is: based on the whole genome genes and their corresponding expression levels or differential expression level information, a preset gene function set that is significantly enriched in the whole genome is found using a statistical model; and / or, the pathway topology analysis PT is: the position of the gene in the pathway, the connectivity with other genes and the regulatory effect information are combined to evaluate the sharing of the gene in the functional gene set and assign corresponding weights, and finally integrate into the enrichment analysis.

8. The drug recommendation method according to claim 7, characterized in that: The perturbation-related database and the database of drug response status or sensitive information include: CMap, LINCS, DrugMatrix, GDSC, CTRP, NCI-60; and / or, The functional modules refer to functional modules including GO function, signaling pathway, transcriptional regulation, and protein-protein interaction.

9. A drug recommendation system for drug efficacy prediction and / or personalized medication guidance, characterized in that: It is based on the drug recommendation method according to any one of claims 1 to 8, wherein the system comprises: a reference database module, a gene differential expression map module, a multi-dimensional differential expression map module, a drug recommendation ranking module and a drug recommendation result module; preferably, The reference database module is used to construct a multi-dimensional differential expression map reference database; The gene differential expression map module is used to obtain gene expression data, and calculate the gene differential expression value under the data to obtain the gene differential expression map; The multi-dimensional differential expression map module is used to calculate the differential expression map under the functional module gene set, and combine the gene differential expression map to obtain a multi-dimensional differential expression map; The drug recommendation ranking module is used to calculate the drug recommendation ranking results under different dimensions and different calculation methods using a similarity statistical analysis method; The drug recommendation result module is used to obtain reliable drug recommendation results using model-integrated statistics for drug efficacy prediction and / or personalized medication guidance.

10. A method for constructing a multi-dimensional differential expression map reference database, characterized in that: The following steps are involved: Step I: Obtaining gene differential expression profiles; Step II: constructing a differential expression map of the functional module gene set: based on the gene differential expression map obtained in step I, the scores in different functional module gene set dimensions are calculated by gene set enrichment analysis to obtain differential expression maps in different functional module dimensions; preferably, the different functional modules include GO functions, signaling pathways, transcriptional regulation, and protein-protein interactions; Step III: Based on the gene differential expression map described in step I and the functional module gene set differential expression map described in step II, a multidimensional differential expression map of genes and functional modules including GO functions, signaling pathways, transcriptional regulation, and protein-protein interactions is constructed, i.e., a multidimensional differential expression map reference database.

11. The method for constructing a multi-dimensional differential expression map reference database according to claim 10, characterized in that: The gene differential expression map includes: a perturbation gene differential expression map and a drug sensitive gene differential expression map; Preferably, the acquisition of the perturbed gene differential expression map comprises: acquiring gene expression data before and after the perturbation from a perturbation-related database and / or literature, calculating the differential expression value of each gene before and after the perturbation, and obtaining the perturbation gene differential expression map; Preferably, the acquisition of the drug-sensitive gene differential expression map comprises: acquiring baseline gene expression data from a database and / or literature containing drug response status or sensitivity information, calculating the differential expression value of each gene in different response states, and obtaining the drug-sensitive gene differential expression map; Preferably, the different response states include: a drug-sensitive group and a drug-insensitive group; Preferably, the functional module gene set differential expression map includes: a perturbation functional module gene set differential expression map and a drug sensitivity functional module gene set differential expression map; The multidimensional differential expression map constructed corresponding to the perturbation gene differential expression map is the perturbation reference database; the multidimensional differential expression map constructed corresponding to the drug sensitivity gene differential expression map is the drug sensitivity reference database; Preferably, the gene expression data is obtained by a gene expression analysis method; preferably, the gene expression analysis method includes: transcriptome detection and analysis, proteome detection and analysis; preferably, the transcriptome detection includes: gene chip, RNA-Seq; Preferably, the method for obtaining the differential expression value of the gene expression data and the corresponding control expression data comprises: logFC and z-score; Preferably, the gene expression data before and after the disturbance refers to the gene expression data of the sample treated with exogenous stimulation and the corresponding control untreated sample; preferably, the exogenous stimulation includes: compound treatment, gene knockout, gene knockdown or gene overexpression, environmental exposure; preferably, the compound includes a mixture; and / or, The gene set enrichment analysis includes over-representation analysis ORA, function set scoring FCS, and pathway topology analysis PT; preferably, the over-representation analysis ORA is: based on a group of genes of interest, a preset gene function set that is significantly enriched in the group of genes is found using a statistical model; and / or, the function set scoring FCS is: based on the whole genome genes and their corresponding expression levels or differential expression level information, a preset gene function set that is significantly enriched in the whole genome is found using a statistical model; and / or, the pathway topology analysis PT is: the position of the gene in the pathway, the connectivity with other genes and the regulatory effect information are combined to evaluate the sharing of the gene in the functional gene set and assign corresponding weights, and finally integrate into the enrichment analysis.

12. The method for constructing a multi-dimensional differential expression map reference database according to claim 11, characterized in that: The perturbation-related database and the database of drug response status or sensitive information include: CMap, LINCS, DrugMatrix, GDSC, CTRP, NCI-60; and / or, The functional modules refer to functional modules including GO function, signaling pathway, transcriptional regulation, and protein-protein interaction.

13. A multi-dimensional differential expression map reference database constructed using the method according to any one of claims 10 to 12.

14. Application of the drug recommendation method as described in any one of claims 1 to 8, the drug recommendation system as described in claim 9, the construction method as described in any one of claims 10 to 12, and the multi-dimensional differential expression map reference database as described in claim 13 in drug efficacy prediction and personalized medication guidance.

15. The use according to claim 14, characterized in that The drug efficacy prediction includes: drug similarity prediction, drug function prediction, which is used for drug reuse or drug mechanism discovery; preferably, the drug efficacy prediction selects a perturbation reference database; and / or, The personalized medication guidance refers to recommending appropriate drugs for different diseases or evaluating the effectiveness of used drugs; preferably, the personalized medication guidance selects a perturbation reference database or a drug sensitivity reference database.

16. A computer device, characterized in that: It includes an input device, a processor, a memory and an output device, wherein the input device, the processor, the memory and the output device are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the method as described in any one of claims 1-8, 10-12.

17. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 and 10 to 12 is implemented.

Images