A genomics-based method for constructing a drug target prediction model for Parkinson's disease

Abstract

The present invention discloses a method for constructing a genomics-based Parkinson's disease drug target prediction model, which relates to the field of drug development technology and includes the following steps: collecting genomic, transcriptomic, and proteomic data related to Parkinson's disease patients and performing quality control and standardization processing; identifying key genes and signaling pathways related to Parkinson's disease through differential expression analysis and functional enrichment, and determining the scope of potential drug targets. By integrating genomic, transcriptomic, and proteomic data from Parkinson's disease patients, the present invention can comprehensively analyze the molecular mechanisms related to Parkinson's disease. Combined with the ant colony algorithm to optimize multi-target combinations, this method effectively overcomes the limitation of traditional single-target models in capturing the comprehensiveness of complex disease networks, significantly improves the accuracy of target prediction, provides more reliable targets for drug development, and reduces the failure rate of clinical trials.

Description

Technical Field

[0001] The present invention relates to the technical field of drug research and development, and in particular to a method for constructing a Parkinson's disease drug target prediction model based on genomics. Background Art

[0002] Parkinson's disease (PD) is a common neurodegenerative disease characterized by movement disorders. Clinical symptoms include tremor, rigidity, bradykinesia, and postural instability. Its main pathological feature is the gradual loss of dopaminergic neurons in the substantia nigra of the brain. Since Parkinson's disease is currently incurable and symptoms gradually worsen as the disease progresses, patients urgently need effective therapeutic drugs to alleviate symptoms and delay disease progression. In the drug development process for Parkinson's disease, the prediction of drug targets is a key step, which can help researchers more efficiently discover possible drug targets and provide guidance for subsequent experimental studies.

[0003] In the existing technology, Parkinson's disease involves multiple molecular mechanisms, and the targets are often diverse and multiple, making it difficult for traditional single-target models to accurately capture all relevant targets, thereby affecting the comprehensiveness and effectiveness of treatment. Therefore, how to combine genomic data with ant colony algorithms to simultaneously optimize multiple drug targets to form a multi-target treatment prediction model to provide guidance for the development of multi-target drugs is the problem to be solved by the present invention. To this end, a genomics-based method for constructing a Parkinson's disease drug target prediction model is proposed. Summary of the Invention

[0004] The present invention aims to provide a method for constructing a Parkinson's disease drug target prediction model based on genomics to solve the problems raised in the above background technology.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for constructing a genomics-based Parkinson's disease drug target prediction model comprises the following steps:

[0007] S1. Collect genomic, transcriptomic, and proteomic data related to Parkinson's disease patients, perform quality control and standardization, and ensure data consistency and usability;

[0008] S2. Identify key genes and signaling pathways associated with Parkinson's disease through differential expression analysis and functional enrichment, and determine the scope of potential drug targets;

[0009] S3. Construct an interaction network containing Parkinson's disease-related genes and potential drug targets, analyze the regulatory relationships between targets to identify key genes and targets;

[0010] S4. Set the parameters of the ant colony algorithm, including the initial value of the pheromone, the heuristic factor, and the volatilization rate, simulate the optimization search process of the target combination, and iteratively search for the optimal target combination based on the ant colony algorithm. Combined with genomic data, the combination efficacy is evaluated to balance the synergistic and antagonistic effects between targets.

[0011] S5. Select the optimal target combination as the basis for a multi-target therapy prediction model, and train the multi-target therapy prediction model by combining known Parkinson's disease target data with a neural network model;

[0012] S6. Perform multi-target prediction based on the multi-target therapy prediction model, screen potential drug targets, and conduct verification to evaluate the effectiveness of the predicted targets.

[0013] A further improvement of the technical solution of the present invention is that: S1 specifically includes:

[0014] Collect genomic, transcriptomic, and proteomic data from Parkinson's disease patients and healthy controls. Standardize the raw data format and perform preliminary screening. Use a quality filtering tool (Trimmomatic) to remove low-quality reads, adapter sequences, and nonspecific signals. Establish a standardized sample information library and record patient clinical information, sequencing platform parameters, and experimental conditions in detail. Genomic data include single-nucleotide polymorphism (SNP) and convolutional neuron (CNV) data, transcriptomic data include RNA-seq data, and proteomic data include protein expression and interaction data.

[0015] For genomic data, the FastQC tool was used to assess sequencing quality and filter low-coverage regions at sites with a depth <10×. Transcriptomics data were quantified using RSEM, outputting FPKM / TPM values ​​and correcting for batch effects using the Combat method. Proteomics data were identified and quantitatively normalized using MaxQuant to unify the dimensions and distribution of different omics data and eliminate technical bias.

[0016] The standardized genomic, transcriptomic, and proteomic data were integrated into a unified analysis platform to establish a multi-omics association matrix. Cross-omics data were aligned through Ensembl ID mapping to verify gene-transcript-protein consistency. Hierarchical clustering was used to evaluate the global structure of the integrated data to ensure that the biological associations between different omics layers were preserved, and ultimately a high-quality, highly consistent multi-omics dataset was output.

[0017] A further improvement of the technical solution of the present invention is that: S2 specifically includes:

[0018] Differential expression analysis was performed on transcriptomic data (RNA-seq) from Parkinson's disease patients and healthy controls. Differentially expressed genes (DEGs) were screened using the DESeq2 tool based on a negative binomial distribution model. A significance threshold was set to identify genes that were significantly upregulated or downregulated in the disease group. Simultaneously, differential analysis was performed on proteomic data to identify proteins with significantly altered expression levels. Cross-omics association analysis (Pearson correlation) was used to integrate transcriptomic and proteomic data to identify gene-protein pairs that synergistically changed at the two omics levels, forming a preliminary screening list of molecules.

[0019] The GO (Gene Ontology) database was used to perform functional annotation on the initial screening molecules, analyzing their enrichment in biological processes, cellular components, and molecular functions. The Kyoto Encyclopedia of Genes and Genomes (KEGG) database was combined to identify significantly enriched signaling pathways. The significance of pathway enrichment was assessed using hypergeometric distribution. Gene set enrichment analysis (GSEA) was then used to verify the overall regulatory trends of key pathways and clarify the core role of differentially expressed molecules in the pathogenesis of Parkinson's disease.

[0020] By integrating differential expression analysis and functional enrichment results, we screened out genes and proteins that had significant expression changes in Parkinson's disease (upregulation of both RNA and protein levels) and were involved in key signaling pathways (PI3K-AKT, MAPK). We combined literature mining and databases (DisGeNET, OMIM) to verify the association between targets and diseases, excluded nonspecific or low-confidence molecules, and ultimately output a list of multiple potential drug targets, covering genes and proteins, and annotated their biological functions, pathway affiliations, and druggability potential.

[0021] A further improvement of the technical solution of the present invention is that: S3 specifically includes:

[0022] We collected Parkinson's disease-associated genes (GWAS, candidate genes screened by transcriptomics) and potential drug targets (differential expression analysis, functional enrichment results), integrated multi-omics data (genomic, transcriptomic, proteomic) and interaction information from public databases (STRING, BioGRID, IntAct) into a standardized format (PSI-MI), covering physical interactions, co-expression, and regulatory relationships, and then constructed a gene-target interaction network, where nodes are genes / proteins and edges are interaction types (direct / indirect).

[0023] Using the Cytoscape tool, we calculated network topology parameters, including degree centrality (number of node connections), betweenness centrality (number of shortest path passes), and closeness centrality (average distance to other nodes), to identify hub genes and key targets in the network. We then used the shortest path algorithm to analyze regulatory relationships between targets, distinguishing between activation, inhibition, or combined regulatory patterns. We also combined time series expression data to reveal key regulatory events in disease progression and identify core nodes with global impact.

[0024] Based on the results of topological analysis, genes ranked in the top 10% by degree centrality and targets with significant betweenness centrality were screened as candidate key molecules. GO / KEGG enrichment analysis was used to verify whether the candidate molecules were enriched in Parkinson's disease-related pathways (p<0.01). Clinical data and experimental evidence were further integrated to evaluate the biological significance and drugability potential of the candidate molecules. Finally, a list of key genes and targets was output, and their network properties and regulatory relationships were annotated.

[0025] A further improvement of the technical solution of the present invention is that the specific process of calculating network topology parameters using the Cytoscape tool includes:

[0026] Genes and proteins were used as nodes and interaction relationships as edges. The node and edge files were imported using the ImportNetwork function of Cytoscape to generate a gene-target interaction network. The node color and edge thickness were set through the Style panel. The network topology parameters including degree centrality, betweenness centrality and closeness centrality were calculated using the Cytoscape plug-in (NetworkAnalyzer). Degree centrality counts the number of edges connected to each node, betweenness centrality calculates the number of times all shortest paths pass through the node, and closeness centrality calculates the average shortest path distance from a node to all other nodes.

[0027] For degree centrality, select the "Degree" indicator in the NetworkAnalyzer plug-in in Cytoscape to generate a node degree value distribution table. For betweenness centrality, select "BetweennessCentrality" in NetworkAnalyzer to identify nodes with betweenness centrality significantly higher than the average. For closeness centrality, select "Closeness Centrality" in NetworkAnalyzer to filter out nodes ranked in the top 5% of closeness centrality.

[0028] The Dijkstra algorithm was used to calculate the shortest path between any two nodes and identify key regulatory nodes on the path. In Cytoscape, the Shortest Path function was used to input the start and target nodes, and the path and intermediate regulatory nodes were output to distinguish regulatory patterns. For activation relationships, the expression levels of nodes in the path were positively correlated. For inhibition relationships, the expression levels of nodes in the path were negatively correlated. For complex regulation, both activation and inhibition relationships existed in the path.

[0029] Time series expression data (RNA-seq data at different disease stages) were mapped to the network, and the temporal expression patterns of the nodes were annotated. A dynamic network model was generated using the DyNet plug-in of Cytoscape. Nodes whose expression levels continued to increase or decrease during disease progression (Fold Change>2 and p<0.01) were screened. Combined with the shortest path analysis, regulatory chains that played a key role in the early or late stages of the disease were identified, and core nodes that regulated multiple pathways or across modules were targeted.

[0030] A further improvement of the technical solution of the present invention is that: S4 specifically includes:

[0031] The core parameters of the ant colony algorithm are set, including the initial pheromone value, heuristic factor, and evaporation rate. The initial pheromone value is set to a small positive value (0.1) with uniform distribution to ensure that the initial path has no preference. The heuristic factor is used to balance the attraction of the local optimal solution and is set to 1 or 2. The evaporation rate is used to simulate the natural evaporation of pheromones to prevent the algorithm from converging prematurely and is set to 0.1 to 0.5. At the same time, the number of ants (20 to 50) and the maximum number of iterations (100 to 500) are set to ensure the algorithm's computational efficiency and convergence. All potential targets are used as nodes to construct a complete graph model.

[0032] The ant colony algorithm is used to simulate the optimization search process of target combinations. Each ant selects the next target based on the pheromone concentration and the probability of the heuristic factor, generates a target combination path, and gradually optimizes the target combination through iterative search. After each round of iteration, the fitness of the current combination is evaluated and the pheromone matrix is ​​updated.

[0033] The efficacy of target combinations is evaluated by combining genomic data, balancing synergistic and antagonistic effects. By constructing synergistic effect scores and antagonistic effect scores, the therapeutic potential of the combination is comprehensively evaluated. The iteration termination condition is set as no change in the optimal solution after 10 consecutive iterations. The target combination corresponding to the path with the highest pheromone concentration is output, and then the recurring core combination is screened through cluster analysis as the candidate optimal solution.

[0034] A further improvement of the technical solution of the present invention is that the analysis and screening process of the candidate optimal solutions is as follows:

[0035] A scoring system is constructed using genomic data to quantify the synergistic and antagonistic effects of target combinations. The synergistic effect score is calculated based on gene co-expression, protein interaction strength, and the number of co-regulated pathways between targets to analyze the synergistic effect between targets. The antagonistic effect score is constructed based on the negative correlation between targets to identify the antagonistic effect between targets.

[0036] A fitness function is constructed by combining the synergistic effect score and the antagonistic effect score to comprehensively evaluate the therapeutic potential of the target combination. In each iteration, the fitness of the current target combination is evaluated and the pheromone matrix is ​​updated based on the fitness. At the same time, the iteration termination condition is set to no change in the optimal solution after 10 consecutive iterations to ensure the convergence of the algorithm.

[0037] Cluster analysis was performed on the target combinations generated in multiple iterations to identify recurring core combinations, and the K-means clustering algorithm was used to group similar target combinations into one category. The effectiveness and reliability of the core combination were further evaluated through cross-validation and experimental verification, and the target combination corresponding to the path with the highest pheromone concentration was output. Combined with the cluster analysis results, the final candidate optimal solution was determined, and the synergistic and antagonistic effect scores, fitness values, and frequency of occurrence in the cluster analysis were annotated in the final output target combination.

[0038] A further improvement of the technical solution of the present invention is that: S5 specifically includes:

[0039] Collect data on the best known Parkinson's disease-related target combinations, integrate multi-dimensional data including gene expression, protein interactions, and clinical information, extract biological features of the targets within the combination, perform Z-score normalization, and generate a structured feature matrix as the basic data for model input;

[0040] A multi-layer feedforward neural network model was constructed, consisting of an input layer, hidden layers, and an output layer. The input layer received a structured feature matrix with dimensions consistent with the number of features. The hidden layer used a two-layer fully connected network with 128 neurons in each layer. The output layer outputted a therapeutic potential score (0-1) using a single neuron. A sigmoid activation function was used, and preclinical Parkinson's disease experimental data was used as labels. A 5-fold cross-validation algorithm was used to divide the training and test sets. The model was iteratively trained until the loss function converged to ensure prediction accuracy.

[0041] The performance of the multi-target treatment prediction model was verified through an independent test set. Evaluation indicators included mean square error (MSE) and receiver operating characteristic curve (ROC-AUC). The SHAP value was combined to analyze the contribution of each target feature to the prediction results. Finally, the trained multi-target treatment prediction model output a treatment potential score, providing a priority ranking scheme for Parkinson's disease drug development, while also annotating potential side effect risks and biological mechanisms.

[0042] A further improvement of the technical solution of the present invention is that the calculation process of the therapeutic potential score is:

[0043] For each pair of targets i and j in the target combination, the ratio of the synergistic effect score to the antagonistic effect score is calculated. The ratio reflects the strength of the synergistic effect relative to the antagonistic effect, and the square root of the ratio of the synergistic effect to the antagonistic effect is calculated;

[0044] Calculate the product of the clinical relevance scores of targets i and j to obtain a comprehensive clinical relevance score, and add 1 to the comprehensive clinical relevance score and take the natural logarithm to obtain the weighted clinical relevance;

[0045] The square root of the ratio of synergistic effect to antagonistic effect is multiplied by the weight of clinical relevance, and then the product of all target pairs (i, j) is double-summed;

[0046] The result of the double summation is substituted into the exponential function, and then the exponential function is added by 1 and the inverse is taken to obtain the final therapeutic potential score. The closer the value of the therapeutic potential score is to 1, the higher the therapeutic potential is, and the closer it is to 0, the lower the therapeutic potential is.

[0047] A further improvement of the technical solution of the present invention is that: S6 specifically includes:

[0048] Based on the trained multi-target therapeutic prediction model, a structured feature matrix of Parkinson's disease-related target combinations was input, and the therapeutic potential score of each combination was output. A score threshold (≥0.7) was set to screen high-potential target combinations. SHAP value analysis was combined to identify key targets within the combination and their biological mechanisms. Through literature mining and database cross-validation, reported low-efficacy or high-toxicity targets were excluded, and a list of candidate targets was generated, with their potential treatment pathways and risks annotated.

[0049] Design multi-level validation experiments for candidate targets, including in vitro experiments, in vivo experiments and mechanism verification;

[0050] Calculate the inhibition rate of target combinations on pathological phenotypes and the degree of improvement in behavioral scores, monitor off-target effects in animal models, assess risks in combination with preclinical toxicology data, output high-confidence target combinations and their priority rankings, annotate biological mechanisms and potential side effects, and provide a list of translatable targets for drug development.

[0051] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0052] 1. The present invention provides a genomics-based method for constructing a Parkinson's disease drug target prediction model. By integrating the genomic, transcriptomic, and proteomic data of Parkinson's patients, it can comprehensively analyze the molecular mechanisms related to Parkinson's disease. Combined with the ant colony algorithm to optimize multi-target combinations, it effectively overcomes the limitation of traditional single-target models that are difficult to capture the comprehensiveness of complex disease networks, significantly improves the accuracy of target prediction, provides more reliable targets for drug development, and reduces the failure rate of clinical trials.

[0053] 2. The present invention provides a method for constructing a genomics-based prediction model for Parkinson's disease drug targets. The constructed prediction model can quickly screen out high-potential multi-target combinations, and further evaluate the therapeutic potential of the targets through multi-dimensional data integration and deep learning model training. This efficient screening and evaluation process significantly shortens the drug development cycle and accelerates the transition from basic research to clinical trials. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0055] Figure 1 Schematic diagram of the workflow of the present invention;

[0056] Figure 2 Schematic diagram of the method of the present invention. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0058] Example 1, as Figure 1 、 Figure 2 As shown, the present invention provides a method for constructing a Parkinson's disease drug target prediction model based on genomics, comprising the following steps:

[0059] S1. Collect genomic, transcriptomic and proteomic data related to Parkinson's disease patients, perform quality control and standardization to ensure data consistency and availability. Collect genomic, transcriptomic and proteomic data from Parkinson's disease patients and healthy controls, unify the format of the original data, preliminarily screen, and then use the quality filtering tool (Trimmomatic) to remove low-quality reads, linker sequences and non-specific signals, establish a standardized sample information library, and record the patient's clinical information (age, gender, disease course, medication history), sequencing platform parameters (sequencing depth, read length) and experimental conditions (sample processing flow, mass spectrometry parameters) in detail. Among them, genomic data include SNP and CNV data, transcriptomic data include RNA-seq data, and proteomic data include protein expression and interaction data. For genomic data, the FastQC tool is used to evaluate sequencing quality, and low coverage regions of sites with a depth <10× are filtered. Transcriptomic data are analyzed by RSEM (RNA-Seq by Expression quantification was performed using the Expectation-Maximization (FPKM / TPM) algorithm, and batch effects were corrected using the Combat method. Proteomics data were identified and quantitatively normalized using MaxQuant to identify peptides. The scale and distribution of different omics data were unified to eliminate technical bias. The standardized genomic, transcriptomic, and proteomic data were integrated into a unified analysis platform. A multi-omics association matrix was established, and cross-omics data were aligned through Ensembl ID mapping to verify gene-transcript-protein consistency. Hierarchical clustering was used to assess the global structure of the integrated data to ensure that biological associations between different omics layers were preserved, ultimately outputting high-quality, highly consistent multi-omics datasets.

[0060] S2. Through differential expression analysis and functional enrichment, key genes and signaling pathways related to Parkinson's disease were identified, and the scope of potential drug targets was determined. Differential expression analysis was performed on transcriptomic data (RNA-seq) of Parkinson's disease patients and healthy controls. Differentially expressed genes (DEGs) were screened using the DESeq2 tool based on the negative binomial distribution model. A significance threshold (|log2FC|>1 and p-adj<0.05) was set to screen out genes that were significantly upregulated or downregulated in the disease group. Simultaneously, differential analysis was performed on proteomic data to identify proteins with significant changes in expression (p<0.05 and Fold Change>1.5). Cross-omics association analysis (Pearson correlation) was used to integrate transcriptomic and proteomic data to screen out gene-protein pairs that changed synergistically at the two omics levels, forming a preliminary screening molecule list. Based on the preliminary screening molecule list, GO (Gene The Kyoto Encyclopedia of Genes and Genomes (KEGG) database was used for functional annotation, and enrichment in biological processes (neurotransmitter metabolism), cellular components (synaptic membrane), and molecular functions (kinase activity) was analyzed (p < 0.01). Significantly enriched signaling pathways were identified using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Pathway enrichment significance was assessed using hypergeometric distribution. Gene set enrichment analysis (GSEA) was then used to validate the overall regulatory trends of key pathways and clarify the core role of differentially expressed molecules in the pathogenesis of Parkinson's disease. The differential expression analysis and functional enrichment results were integrated to screen for genes and proteins with significant expression changes (upregulation at both RNA and protein levels) in Parkinson's disease and involved in key signaling pathways (PI3K-AKT, MAPK). The association between targets and disease was verified by combining literature mining and databases (DisGeNET and OMIM), and nonspecific or low-confidence molecules were excluded. Finally, a list of multiple potential drug targets was generated, covering genes and proteins, with annotations of their biological functions, pathway affiliations, and druggability potential.

[0061] S3. Construct an interaction network including Parkinson's disease-related genes and potential drug targets, analyze the regulatory relationship between targets to identify key genes and targets, collect Parkinson's disease-related genes (GWAS, candidate genes screened by transcriptomics) and potential drug targets (differential expression analysis, functional enrichment results), integrate multi-omics data (genome, transcriptome, proteome) and interaction information from public databases (STRING, BioGRID, IntAct), unify them into a standardized format (PSI-MI), covering physical interactions, co-expression and regulatory relationships, and then construct a gene-target interaction network, where nodes are genes / proteins and edges are interaction types (direct / indirect). Use Cytoscape tools to calculate network topology parameters, including degree centrality (number of node connections) and degree centrality (number of node connections). ), betweenness centrality (number of shortest path passes) and closeness centrality (average distance to other nodes), identify hub genes and key targets in the network, analyze the regulatory relationships between targets using the shortest path algorithm, distinguish activation, inhibition or compound regulatory patterns, and combine time series expression data to reveal key regulatory events in disease progression, identify core nodes with global impact, and based on the topological analysis results, screen the top 10% genes in degree centrality and targets with significant betweenness centrality as candidate key molecules. GO / KEGG enrichment analysis verifies whether the candidate molecules are enriched in Parkinson's disease-related pathways (p<0.01). Further integration of clinical data and experimental evidence evaluates the biological significance and druggability potential of the candidate molecules, and ultimately outputs a list of key genes and targets, annotating their network properties and regulatory relationships;

[0062] In addition, the specific process of calculating network topology parameters using the Cytoscape tool includes:

[0063] Genes and proteins are taken as nodes and interaction relationships as edges. The node and edge files are imported using the ImportNetwork function of Cytoscape to generate a gene-target interaction network. The node color and edge thickness are set through the Style panel. The network topology parameters including degree centrality, betweenness centrality and closeness centrality are calculated using the Cytoscape plug-in (NetworkAnalyzer). Degree centrality counts the number of connected edges of each node, reflecting the direct interaction ability of the node in the network. Betweenness centrality calculates the number of times all shortest paths pass through the node, reflecting the bridge role of the node in information transmission. Closeness centrality calculates the average shortest path distance from the node to all other nodes, reflecting the global reachability of the node in the network. For degree centrality, select the "Degree" indicator in the NetworkAnalyzer plug-in in Cytoscape to generate a node degree value distribution table. Nodes with high degree centrality (top 10% degree value) are usually hub genes and may be involved in multi-pathway regulation. For betweenness centrality, select "Betweenness Centrality" in NetworkAnalyzer. "Centrality" identifies nodes with betweenness centrality significantly higher than the average. Nodes with high betweenness centrality (betweenness value > 0.1) may be key targets, and their abnormal regulation may cause global network collapse. For closeness centrality, select "Closeness Centrality" in Network Analyzer to screen the top 5% nodes in closeness centrality. Nodes with high closeness centrality may respond quickly to signal changes and play an early regulatory role in disease progression. The Dijkstra algorithm is used to calculate the shortest path between any two nodes and identify key regulatory nodes on the path. In Cytoscape, the Shortest Path function is used to input the start node and target node, output the path and intermediate regulatory nodes, and then distinguish the regulatory patterns. For activation relationships, the expression levels of nodes in the path are positively correlated, for inhibition relationships, the expression levels of nodes in the path are negatively correlated, and for complex regulation, both activation and inhibition relationships exist in the path. Time series expression data (RNA-seq data at different disease stages) are mapped to the network, and the temporal expression patterns of nodes are annotated. Using Cytoscape's DyNet plug-in, a dynamic network model is generated to screen nodes whose expression levels continuously increase or decrease during disease progression (Fold Change>2 and p<0.01), combined with the shortest path analysis, to identify the regulatory chains that play a key role in the early or late stages of the disease and to identify the core nodes that regulate multiple pathways or cross-modules;

[0064] S4. Set the parameters of the ant colony algorithm, including the initial value of the pheromone, the heuristic factor, and the volatilization rate, simulate the optimization search process of the target combination, and iteratively search for the optimal target combination based on the ant colony algorithm. Combined with genomic data, the combination efficacy is evaluated to balance the synergistic and antagonistic effects between targets.

[0065] S5. Select the optimal target combination as the basis for a multi-target therapy prediction model, and train the multi-target therapy prediction model by combining known Parkinson's disease target data with a neural network model;

[0066] S6. Perform multi-target prediction based on the multi-target therapy prediction model, screen potential drug targets, and conduct verification to evaluate the effectiveness of the predicted targets.

[0067] Example 2, as Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, S4 specifically includes:

[0068] The core parameters of the ant colony algorithm are set, including the initial value of pheromone, the heuristic factor and the volatilization rate. The initial value of pheromone is set to a small positive value (0.1) with uniform distribution to ensure that the initial path has no preference. The heuristic factor is used to balance the attraction of the local optimal solution and is set to 1 or 2. The volatilization rate is used to simulate the natural volatilization of pheromone to prevent the algorithm from converging prematurely and is set to 0.1 to 0.5. At the same time, the number of ants (20 to 50) and the maximum number of iterations (100 to 500) are set to ensure the computational efficiency and convergence of the algorithm. All potential targets are taken as nodes to construct a complete graph model. The edge weights are determined by the heuristic factor. The pheromone matrix is ​​initialized and the number of ants is set to 10%-20% of the total number of targets. The ant colony algorithm is used to simulate the optimization search process of the target combination. Each ant selects the target based on the pheromone concentration and the heuristic factor. The next target is selected based on the probability of the trigger factor, and a target combination path is generated. The target combination is gradually optimized through iterative search. After each round of iteration, the fitness of the current combination is evaluated and the pheromone matrix is ​​updated. The heuristic information can be combined with the target correlation in the genomic data to guide the ants to search for high-potential areas. The pheromone update rule is as follows: pheromone is added to the high-quality target combination path and reduced on the poor path. The efficacy of the target combination is evaluated in combination with genomic data, and the synergistic and antagonistic effects are balanced. By constructing synergistic effect scores and antagonistic effect scores, the therapeutic potential of the combination is comprehensively evaluated. The optimal solution is set to remain unchanged for 10 consecutive iterations as the iteration termination condition. The target combination corresponding to the path with the highest pheromone concentration is output, and the recurring core combination is screened through cluster analysis as the candidate optimal solution.

[0069] The expression for generating the target combination pathway is:

[0070]

[0071] Where R ij is the probability of the ant moving from target i to target j, which determines the next target selection. The higher the probability value, the greater the possibility of choosing this path. ij is the pheromone concentration on the path from target point i to target point j, reflecting the historical attractiveness of the path. The higher the pheromone concentration, the greater the probability of the path being selected. ij is the heuristic factor from target i to target j), reflecting the prior advantage of the path. The higher the heuristic factor, the greater the probability of the path being selected. z1 is the pheromone importance parameter (usually set to 1), which controls the influence weight of pheromone on path selection. The larger z1 is, the more significant the effect of pheromone is. z2 is the heuristic factor importance parameter (usually set to 1 or 2), which controls the influence weight of heuristic factor on path selection. The larger z2 is, the more significant the effect of heuristic factor is. allowed is the set of targets currently accessible to ants (targets that have not been visited), ensuring that ants will not repeatedly visit selected nodes to avoid path cycles. is the z1 power of the pheromone concentration, amplifying the attraction of high pheromone paths, is the z2 power of the inspiration factor, which amplifies the attractiveness of high inspiration factor paths. To sum the combined attractiveness of all accessible targets k, normalize the probability distribution to ensure that all R ij The sum is 1;

[0072] In addition, the analysis and screening process of candidate optimal solutions is:

[0073] A scoring system is constructed through genomic data to quantify the synergistic and antagonistic effects of target combinations. For the synergistic effect score, the synergistic effect between targets is analyzed based on the gene co-expression, protein interaction strength, and number of co-regulatory pathways between targets. For the antagonistic effect score, the negative correlation between targets is constructed to identify the antagonistic effect between targets. The fitness function is constructed by combining the synergistic effect score and the antagonistic effect score to comprehensively evaluate the therapeutic potential of the target combination. In each round of iteration, the fitness of the current target combination is evaluated, and the pheromone matrix is ​​updated according to the fitness. At the same time, a continuous 1 The iteration termination condition is that the optimal solution of 0 iterations remains unchanged to ensure the convergence of the algorithm. Cluster analysis is performed on the target combinations generated in multiple iterations to identify repeated core combinations. The K-means clustering algorithm is used to group similar target combinations into one category. The effectiveness and reliability of the core combination are further evaluated through cross-validation and experimental verification. The target combination corresponding to the path with the highest pheromone concentration is output. Combined with the cluster analysis results, the final candidate optimal solution is determined. The synergistic and antagonistic effect scores, fitness values, and frequency of occurrence in the cluster analysis are annotated in the final output target combination.

[0074] The calculation expression of the synergy effect score is:

[0075]

[0076] The calculation expression of the antagonistic effect score is:

[0077]

[0078] The calculation expression of the fitness function is:

[0079]

[0080] Where S ij D is the synergistic effect score, which indicates the synergistic effect strength between targets. ij is the antagonistic effect score, which indicates the intensity of the antagonistic effect between targets. F is the fitness function value, which comprehensively evaluates the therapeutic potential of the target combination. n is the number of targets in the target combination. G ij is the gene co-expression correlation between target i and target j, P ij is the protein interaction strength between target i and target j, K ij is the number of co-regulatory pathways between target i and target j, C ij Score the clinical relevance of target i and target j, N ij is the negative correlation strength between target i and target j, L ij is the distance between target i and target j, α and β are weight parameters used to balance the effects of synergistic and antagonistic effects. It was determined through cross-validation that the synergistic effect score increases with the increase of gene co-expression, protein interaction strength and the number of co-regulated pathways, and the antagonistic effect score increases with the increase of negative correlation strength. The fitness function value increases with the increase of synergistic effect and the decrease of antagonistic effect. The closer the value is to 1, the higher the therapeutic potential.

[0081] Example 3, as Figure 1 、 Figure 2 As shown, based on Examples 1-2, the present invention provides a technical solution: preferably, S5 specifically includes:

[0082] Collect the best target combination data related to Parkinson's disease, integrate multi-dimensional data including gene expression, protein interaction and clinical information, extract biological characteristics of the targets in the combination, and perform Z-score normalization to generate a structured feature matrix as the basic data for model input. Among them, gene expression data is used to obtain the expression spectrum of the target in different tissues or cell types and quantify its dynamic regulatory characteristics. Protein interaction data is used to extract the physical interaction and functional association strength between targets based on the STRING or BioGRID database. Clinical information is used to integrate the correlation between the target and the Parkinson's disease phenotype. A multi-layer feedforward neural network model is constructed, including an input layer, a hidden layer and an output layer. The input layer is used to receive the structured feature matrix with the same dimension as the number of features. The hidden layer uses a two-layer fully connected network with 128 neurons in each layer. The output layer outputs a therapeutic potential score (0-1) through a single neuron, using a Sigmoid activation function and Dropout regularization to prevent overfitting. The Adam optimizer is used to adjust the weights. Preclinical experimental data for Parkinson's disease are used as labels. The training set and test set are divided through 5-fold cross-validation. The model is iteratively trained until the loss function converges to ensure prediction accuracy. The performance of the multi-target therapeutic prediction model is verified by an independent test set. Evaluation indicators include mean square error (MSE) and receiver operating characteristic curve (ROC-AUC). The SHAP value is combined to analyze the contribution of each target feature to the prediction result. Finally, the trained multi-target therapeutic prediction model outputs a therapeutic potential score, providing a priority sorting scheme for Parkinson's disease drug development, while also annotating potential side effect risks and biological mechanisms.

[0083] In addition, the calculation process of the therapeutic potential score is:

[0084] For each pair of targets i and j in the target combination, the ratio of the synergistic effect score to the antagonistic effect score is calculated. The ratio reflects the strength of the synergistic effect relative to the antagonistic effect, and the square root of the ratio of the synergistic effect to the antagonistic effect is calculated. The square root operation is used to balance the effects of the synergistic effect and the antagonistic effect, so that when the synergistic effect is much greater than the antagonistic effect, the value is larger, and vice versa. The product of the clinical relevance scores of targets i and j is calculated to obtain a comprehensive clinical relevance score, and the natural logarithm of the comprehensive clinical relevance score plus 1 is taken. The natural logarithm operation is used to adjust the weight of the synergistic effect so that the target pairs with higher clinical relevance contribute more to the overall therapeutic potential. Then, the weight of clinical relevance is obtained, and the square root of the ratio of synergistic effect to antagonistic effect is multiplied by the weight of clinical relevance. Then, the product results of all target pairs (i, j) are doubly summed. The doubly summed result ensures that the interactions of all target pairs are taken into account, and comprehensively reflects the synergistic and antagonistic effects of the entire target combination. The result of the doubly summed result is substituted into the exponential function. The exponential function maps the result of the linear combination to an S-shaped curve, making the result easier to interpret and compare. Then, the exponential function is added by 1 and the inverse is taken to obtain the final therapeutic potential score. The closer the value of the therapeutic potential score is to 1, the higher the therapeutic potential is, and the closer it is to 0, the lower the therapeutic potential is.

[0085] The calculation expression of the therapeutic potential score is:

[0086]

[0087] Where TPS is the therapeutic potential score, which indicates the probability of successful treatment, n is the number of targets in the target combination, S ij The synergistic effect score between target i and target j ranges from 0 to 1. The larger the value, the stronger the synergistic effect. ij The antagonistic effect score between target i and target j ranges from 0 to 1. The larger the value, the stronger the antagonistic effect. i C is the clinical relevance score of target i, ranging from 0 to 1. The larger the value, the stronger the correlation between the target and the Parkinson's disease phenotype. j is the clinical relevance score of target j, ranging from 0 to 1. A larger value indicates a stronger correlation between the target and the Parkinson's disease phenotype. The TPS value ranges from 0 to 1. The closer the value is to 1, the higher the therapeutic potential is, and the closer the value is to 0, the lower the therapeutic potential is. When the synergistic effect score increases, the TPS value increases. When the antagonistic effect score increases, the TPS value decreases. When the clinical relevance score of the target increases, the TPS value increases.

[0088] S6 specifically includes:

[0089] Based on the trained multi-target therapeutic prediction model, the structured feature matrix of Parkinson's disease-related target combinations was input, and the therapeutic potential score of each combination was output. A score threshold (≥0.7) was set to screen high-potential target combinations. The key targets in the combination and their biological mechanisms were identified by combining SHAP value analysis. Through literature mining and database cross-validation, reported low-efficiency or high-toxicity targets were excluded, and a list of candidate targets was generated. Their potential treatment paths and risks were marked. Multi-level verification experiments were designed for the candidate targets, including in vitro experiments, in vivo experiments and mechanism verification. Among them, the in vitro experiments used cell models (dopaminergic neurons differentiated from induced pluripotent stem cells) to evaluate the effect of target regulation on α-synapses. The effects of target combinations on nuclear protein aggregation and mitochondrial function were evaluated in vivo using an animal model of Parkinson's disease (MPTP-induced mouse model) to verify the improvement effect of target combinations on motor behavior and neuropathological changes. The mechanism was verified using RNA interference and CRISPR knockout technology to confirm target function, and proteomics was combined to analyze changes in downstream pathways to ensure that the target mechanism of action was consistent with the prediction. The inhibition rate of the target combination on the pathological phenotype and the improvement in behavioral scores were calculated, and off-target effects in animal models were monitored. Preclinical toxicology data were combined to assess risks, output high-confidence target combinations and their priority rankings, annotate biological mechanisms and potential side effects, and provide a list of transformable targets for drug development.

[0090] The above are only specific embodiments of the present application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for constructing a drug target prediction model for Parkinson's disease based on genomics, characterized in that: The following steps are involved: S1. Collect genomic, transcriptomic, and proteomic data related to Parkinson's disease patients and perform quality control and standardization. S2. Identify key genes and signaling pathways associated with Parkinson's disease through differential expression analysis and functional enrichment, and determine the scope of potential drug targets; S3. Construct an interaction network containing Parkinson's disease-related genes and potential drug targets, analyze the regulatory relationships between targets to identify key genes and targets; S4. Set the parameters of the ant colony algorithm and iteratively search for the optimal target combination based on the ant colony algorithm. Combined with genomic data, the efficacy of the combination is evaluated to balance the synergistic and antagonistic effects between targets. S5. Select the optimal target combination as the basis for a multi-target therapy prediction model, and train the multi-target therapy prediction model by combining known Parkinson's disease target data with a neural network model; S6. Perform multi-target prediction based on the multi-target therapy prediction model, screen potential drug targets, and conduct verification to evaluate the effectiveness of the predicted targets.

2. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 1, characterized in that: Said S1 specifically includes: Collect genomic, transcriptomic, and proteomic data from Parkinson's disease patients and healthy controls. Standardize the format of the raw data, perform preliminary screening, and use quality filtering tools to remove low-quality reads, adapter sequences, and nonspecific signals. Establish a standardized sample information library and record patient clinical information, sequencing platform parameters, and experimental conditions in detail. Genomic data includes single-nucleotide polymorphism (SNP) and convolutional neuron (CNV) data, transcriptomic data includes RNA-seq data, and proteomic data includes protein expression and interaction data. For genomic data, the FastQC tool was used to assess sequencing quality and filter low-coverage regions. Transcriptomics data were quantified using RSEM, with FPKM / TPM values ​​output and batch effects corrected using the Combat method. Proteomics data were identified and quantitatively normalized using MaxQuant to unify the dimensions and distribution of different omics data. The standardized genomic, transcriptomic, and proteomic data were integrated into a unified analysis platform to establish a multi-omics association matrix. Cross-omics data were aligned through Ensembl ID mapping to verify gene-transcript-protein consistency. Hierarchical clustering was used to evaluate the global structure of the integrated data, ultimately outputting high-quality, highly consistent multi-omics datasets.

3. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 1, characterized in that: The S2 specifically includes: Differential expression analysis was performed on transcriptomic data from Parkinson's disease patients and healthy controls. The DESeq2 tool was used to screen for differentially expressed genes based on a negative binomial distribution model. A significance threshold was set to identify genes that were significantly upregulated or downregulated in the disease group. Simultaneously, differential analysis was performed on proteomic data to identify proteins with significantly altered expression levels. Cross-omics association analysis was then performed to integrate transcriptomic and proteomic data, identifying gene-protein pairs that synergistically changed at both omics levels to form a preliminary screening list of molecules. For the initial list of molecules, functional annotation was performed using the GO database to analyze their enrichment in biological processes, cellular components, and molecular functions. The KEGG database was combined to identify significantly enriched signaling pathways. The significance of pathway enrichment was assessed using hypergeometric distribution. Gene set enrichment analysis was then used to verify the overall regulatory trends of key pathways and clarify the core role of differentially expressed molecules in the pathological mechanisms of Parkinson's disease. By integrating differential expression analysis and functional enrichment results, we screen out genes and proteins that have significant expression changes in Parkinson's disease and are involved in key signaling pathways. We then combine literature mining and database analysis to verify the association between targets and the disease, exclude non-specific or low-confidence molecules, and ultimately output a list of multiple potential drug targets, covering genes and proteins, and annotating their biological functions, pathway affiliations, and druggability potential.

4. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 1, characterized in that: The S3 specifically includes: Collect Parkinson's disease-related genes and potential drug targets, integrate multi-omics data with interaction information from public databases, and then construct a gene-target interaction network, where nodes are genes / proteins and edges are interaction types; Using the Cytoscape tool to calculate network topology parameters, including degree centrality, betweenness centrality, and closeness centrality, we identified hub genes and key targets in the network. Using the shortest path algorithm, we analyzed the regulatory relationships between targets, distinguishing activation, inhibition, or complex regulatory patterns. Furthermore, we combined time series expression data to reveal key regulatory events in disease progression and identify core nodes with global impact. Based on the results of topological analysis, the top 10% genes in degree centrality and targets with significant betweenness centrality were screened as candidate key molecules. GO / KEGG enrichment analysis was used to verify whether the candidate molecules were enriched in Parkinson's disease-related pathways. Clinical data and experimental evidence were further integrated to evaluate the biological significance and drugability potential of the candidate molecules. Finally, a list of key genes and targets was output, and their network properties and regulatory relationships were annotated.

5. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 4, characterized in that: The specific process of calculating network topology parameters using the Cytoscape tool includes: Genes and proteins were used as nodes and interaction relationships as edges. The node and edge files were imported using the Import Network function of Cytoscape to generate a gene-target interaction network. The node color and edge thickness were set through the Style panel. The network topology parameters including degree centrality, betweenness centrality, and closeness centrality were calculated using the Cytoscape plug-in. Degree centrality counts the number of edges connected to each node, betweenness centrality calculates the number of times all shortest paths pass through the node, and closeness centrality calculates the average shortest path distance from a node to all other nodes. For degree centrality, select the "Degree" metric in the NetworkAnalyzer plugin in Cytoscape to generate a node degree distribution table. For betweenness centrality, select "Betweenness Centrality" in NetworkAnalyzer to identify nodes with betweenness centrality significantly above the average. For closeness centrality, select "Closeness Centrality" in NetworkAnalyzer to filter out nodes that rank in the top 5% in closeness centrality. The Dijkstra algorithm was used to calculate the shortest path between any two nodes and identify key regulatory nodes on the path. In Cytoscape, the Shortest Path function was used to input the start and target nodes, and the path and intermediate regulatory nodes were output to distinguish regulatory patterns. For activation relationships, the expression levels of nodes in the path were positively correlated. For inhibition relationships, the expression levels of nodes in the path were negatively correlated. For complex regulation, both activation and inhibition relationships existed in the path. Map time series expression data to the network, annotate the temporal expression patterns of nodes, and use Cytoscape's DyNet plug-in to generate a dynamic network model. Filter nodes whose expression levels continuously increase or decrease during disease progression. Combined with shortest path analysis, identify regulatory chains that play a key role in the early or late stages of the disease and lock in core nodes that regulate multiple pathways or across modules.

6. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 1, characterized in that: The S4 specifically includes: The core parameters of the ant colony algorithm are set, including the initial pheromone value, heuristic factor, and evaporation rate. The initial pheromone value is set to a small positive value with uniform distribution. The heuristic factor is used to balance the attraction of the local optimal solution and is set to 1 or 2. The evaporation rate is used to simulate the natural evaporation of pheromones and is set to 0.1 to 0.

5. At the same time, the number of ants and the maximum number of iterations are set. All potential targets are used as nodes to construct a complete graph model. The ant colony algorithm is used to simulate the optimization search process of target combinations. Each ant selects the next target based on the pheromone concentration and the probability of the heuristic factor, generates a target combination path, and gradually optimizes the target combination through iterative search. After each round of iteration, the fitness of the current combination is evaluated and the pheromone matrix is ​​updated. The efficacy of target combinations is evaluated by combining genomic data, balancing synergistic and antagonistic effects. By constructing synergistic effect scores and antagonistic effect scores, the therapeutic potential of the combination is comprehensively evaluated. The iteration termination condition is set as no change in the optimal solution after 10 consecutive iterations. The target combination corresponding to the path with the highest pheromone concentration is output, and then the recurring core combination is screened through cluster analysis as the candidate optimal solution.

7. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 6, characterized in that: The analysis and screening process of the candidate optimal solution is as follows: A scoring system is constructed using genomic data to quantify the synergistic and antagonistic effects of target combinations. The synergistic effect score is calculated based on gene co-expression, protein interaction strength, and the number of co-regulated pathways between targets to analyze the synergistic effect between targets. The antagonistic effect score is constructed based on the negative correlation between targets to identify the antagonistic effect between targets. A fitness function is constructed by combining the synergistic effect score and the antagonistic effect score to comprehensively evaluate the therapeutic potential of the target combination. In each iteration, the fitness of the current target combination is evaluated and the pheromone matrix is ​​updated based on the fitness. At the same time, the iteration termination condition is set as no change in the optimal solution after 10 consecutive iterations. Cluster analysis was performed on the target combinations generated in multiple iterations to identify recurring core combinations, and the K-means clustering algorithm was used to group similar target combinations into one category. The effectiveness and reliability of the core combination were further evaluated through cross-validation and experimental verification, and the target combination corresponding to the path with the highest pheromone concentration was output. Combined with the cluster analysis results, the final candidate optimal solution was determined, and the synergistic and antagonistic effect scores, fitness values, and frequency of occurrence in the cluster analysis were annotated in the final output target combination.

8. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 1, characterized in that: The S5 specifically includes: Collect data on the best known Parkinson's disease-related target combinations, integrate multi-dimensional data including gene expression, protein interactions, and clinical information, extract biological features of the targets within the combination, perform Z-score normalization, and generate a structured feature matrix as the basic data for model input; A multi-layer feedforward neural network model was constructed, consisting of an input layer, hidden layers, and an output layer. The input layer received a structured feature matrix. The hidden layers used a two-layer fully connected network with 128 neurons in each layer. The output layer outputted a therapeutic potential score using a single neuron. A sigmoid activation function was used, and preclinical Parkinson's disease data was used as labels. The model was divided into training and test sets using 5-fold cross-validation, and iterative training was performed until the loss function converged. The performance of the multi-target treatment prediction model was verified through an independent test set. Evaluation indicators included mean square error and receiver operating characteristic curve. The SHAP value was combined to analyze the contribution of each target feature to the prediction results. Finally, the trained multi-target treatment prediction model output a treatment potential score, providing a priority ranking scheme for Parkinson's disease drug development, while also annotating potential side effect risks and biological mechanisms.

9. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 8, characterized in that: The calculation process of the therapeutic potential score is: For each pair of targets i and j in the target combination, the ratio of the synergistic effect score to the antagonistic effect score is calculated, and the square root of the ratio of the synergistic effect to the antagonistic effect is calculated; Calculate the product of the clinical relevance scores of targets i and j to obtain a comprehensive clinical relevance score, and add 1 to the comprehensive clinical relevance score and take the natural logarithm to obtain the weighted clinical relevance; The square root of the ratio of synergistic effect to antagonistic effect is multiplied by the weight of clinical relevance, and then the product of all target pairs (i, j) is double-summed; The result of the double summation is substituted into the exponential function, and then the exponential function is added by 1 and the reciprocal is taken to obtain the final therapeutic potential score.

10. The method for constructing a genomics-based Parkinson's disease drug target prediction model according to claim 9, characterized in that: The S6 specifically includes: Based on the trained multi-target therapeutic prediction model, a structured feature matrix of Parkinson's disease-related target combinations was input, and the therapeutic potential score of each combination was output. A scoring threshold was set to screen high-potential target combinations. SHAP value analysis was combined to identify key targets within the combination and their biological mechanisms. Through literature mining and database cross-validation, reported low-efficacy or high-toxicity targets were excluded, and a list of candidate targets was generated, with their potential treatment pathways and risks annotated. Design multi-level validation experiments for candidate targets, including in vitro experiments, in vivo experiments and mechanism verification; Calculate the inhibition rate of target combinations on pathological phenotypes and the degree of improvement in behavioral scores, monitor off-target effects in animal models, assess risks in combination with preclinical toxicology data, output high-confidence target combinations and their priority rankings, annotate biological mechanisms and potential side effects, and provide a list of translatable targets for drug development.

Images