CN-122025207-A - Drug collaborative screening system based on artificial intelligence prediction and metabonomics verification

CN122025207ACN 122025207 ACN122025207 ACN 122025207ACN-122025207-A

Abstract

The invention discloses a drug collaborative screening system based on artificial intelligence prediction and metabonomics verification, which belongs to the technical field of crossing of artificial intelligence and bioinformatics and comprises a multi-source data acquisition and feature fusion module, an artificial intelligence prediction module, a drug screening and visualization module, a metabonomics verification module and a feedback optimization and self-learning module; the invention realizes unified modeling of a medicine-disease-molecular network through multi-mode data integration and feature coding, realizes medicine combination synergistic effect prediction through an artificial intelligent algorithm, verifies a prediction result by utilizing non-targeted metabonomics data, realizes mechanism interpretability enhancement of a model, builds a technical system capable of closed-loop optimization, and provides an algorithm and experiment double verification platform for medicine discovery and accurate medicine.

Inventors

ZHAO ZHENYING
YANG FENGKUN
DONG LINYI

Assignees

天津市人民医院
天津医科大学

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (10)

1. The drug collaborative screening system based on artificial intelligent prediction and metabonomics verification is characterized by comprising a multi-source data acquisition and feature fusion module, an artificial intelligent prediction module, a drug screening and visualization module, a metabonomics verification module and a feedback optimization and self-learning module; the output end of the multi-source data acquisition and characteristic fusion module is connected with the input end of the artificial intelligent prediction module and is used for acquiring multi-source data and performing characteristic processing to construct a heterogeneous network; the output end of the artificial intelligent prediction module is connected with the input end of the drug screening and visualizing module, and drug pair synergy scores are output based on the input heterogeneous network data; The output end of the drug screening and visualizing module is connected with the input end of the metabonomics verification module, candidate drug combinations are screened according to the cooperative scores, a sample after the prediction combination treatment is obtained, and visual display is carried out; the output end of the metabonomics verification module is connected with the input end of the feedback optimization and self-learning module, and metabonomics analysis is carried out on the candidate medicine combination to obtain a verification result; and the feedback optimization and self-learning module optimizes model parameters based on the verification result and feeds the optimized parameters back to the artificial intelligent prediction module.
2. The artificial intelligence prediction and metabonomics verification-based drug co-screening system of claim 1, wherein the multi-source data acquisition and feature fusion module comprises: Collecting disease gene expression, drug targets and structural information; Carrying out graph structure coding on the drug characteristics, and representing the target characteristics in an One-Hot or Word2Vec mode; the drug-disease-gene three-layer heterogeneous network is constructed and output in the form of an adjacency matrix or tensor.
3. The drug collaborative screening system based on artificial intelligence prediction and metabonomics verification according to claim 1, wherein the multi-source data collection and feature fusion module further comprises a step of jointly embedding drug chemical structures and target gene features to form a drug multidimensional expression vector, wherein the formula is: In the formula, Representing the vector for a multi-dimensional feature of the drug; representing a structural feature vector obtained from a pharmaceutical chemistry structure SMILES sequence via an embedded network; representing a target feature vector obtained from drug target gene/protein features via an embedded network; representing the operation of stitching two vectors in the feature dimension.
4. The drug collaborative screening system based on artificial intelligence prediction and metabonomics verification according to claim 1, wherein the artificial intelligence prediction module obtains an AI prediction result by adopting an AI prediction model with a multi-task learning structure, wherein a primary task is drug pair collaborative score prediction, and a secondary task is target path correlation identification.
5. The artificial intelligence prediction and metabonomics verification-based drug collaborative screening system of claim 4, wherein the AI prediction model further includes computing drug-drug potential effects via a dual channel attention mechanism, the drug-to-feature interaction mapping formula: ; In the middle of The interaction feature vector representing the drug pair is used for describing the action relationship between two drugs; A multi-dimensional feature representation vector representing drug 1; a multi-dimensional feature representation vector representing drug 2; the feature interaction mapping function based on the attention mechanism is expressed and is used for calculating the weighted fusion result of the two drug features under interaction.
6. The artificial intelligence prediction and metabonomics verification-based drug co-screening system of claim 4, wherein the optimized objective function of the AI prediction model is: Wherein, the In order to predict the error term(s), Is used for keeping the consistency of the medicine-target point, Representing the structural constraints of the biological pathway, Weight coefficient as consistency constraint term for adjusting The degree of impact on overall loss; weight coefficients for path structure constraints for controlling Contribution size in model training.
7. The artificial intelligence prediction and metabonomics verification-based drug co-screening system of claim 1, wherein the drug screening and visualization module comprises: Screening candidate drug combinations according to the cooperative score and a preset confidence interval threshold; and displaying the network topological relation and the access target coverage of the medicine through a visual interface.
8. The artificial intelligence prediction and metabonomics verification-based drug co-screening system of claim 4, wherein the metabonomics verification module comprises: adopting an LC-MS/MS non-targeted metabonomics platform to carry out metabolic spectrum analysis on the sample after the prediction combination treatment; Extracting metabolite difference characteristics through PCA, PLS-DA and OPLS-DA algorithms; Pathway enrichment analysis was performed using KEGG and MetaboAnalyst to determine metabolic pathways consistent with AI predictions.
9. The drug collaborative screening system based on artificial intelligence prediction and metabonomics verification according to claim 1, wherein the feedback optimization and self-learning module introduces a path regularization term in a model training process, optimizes AI model weights by adopting a transfer learning or parameter updating mechanism, and feeds the optimized AI model weights back to the artificial intelligence prediction module.
10. The artificial intelligence prediction and metabonomics verification-based drug co-screening system of claim 9, wherein the path regularization term is formulated as: Wherein, the The path correlation of AI predictions is represented, Representing the significance of the differences in metabonomics-validated pathways, Represent the first The weight coefficient of each path is used for measuring the importance degree of the path in the calculation of the regularization term.

Description

Drug collaborative screening system based on artificial intelligence prediction and metabonomics verification Technical Field The invention relates to the technical field of crossing of artificial intelligence and bioinformatics, in particular to a drug collaborative screening system based on artificial intelligence prediction and metabonomics verification. Background The existing drug combination screening mainly has the following technical bottlenecks: Data isolation, namely distributing drug targets, chemical structures and disease molecule expression data in different databases, and lacking unified feature expression and associated modeling means; the algorithm has the limitations that the traditional neural network or random forest model only predicts based on single-source drug characteristics, and does not combine with disease multiple-study information, so that the collaborative prediction accuracy is low; Verification is lacking, namely the existing AI prediction result often lacks molecular level verification, and the interpretable relation between the algorithm result and the biological mechanism is difficult to establish; and the flow is decentralized, namely, a system architecture is not unified from data acquisition to mechanism verification, and quick screening and feedback iteration cannot be realized. Therefore, how to provide a drug co-screening system based on artificial intelligence prediction and metabonomics verification is a problem that needs to be solved by those skilled in the art. Disclosure of Invention In view of the above, the invention provides a drug collaborative screening system based on artificial intelligence prediction and metabonomics verification, which is used for solving the technical problems existing in the prior art and realizing rapid screening, verification and feedback optimization of drug collaborative effects. In order to achieve the above purpose, the present invention adopts the following technical scheme: A drug collaborative screening system based on artificial intelligent prediction and metabonomics verification comprises a multi-source data acquisition and feature fusion module, an artificial intelligent prediction module, a drug screening and visualization module, a metabonomics verification module and a feedback optimization and self-learning module; the output end of the multi-source data acquisition and characteristic fusion module is connected with the input end of the artificial intelligent prediction module and is used for acquiring multi-source data and performing characteristic processing to construct a heterogeneous network; the output end of the artificial intelligent prediction module is connected with the input end of the drug screening and visualizing module, and drug pair synergy scores are output based on the input heterogeneous network data; The output end of the drug screening and visualizing module is connected with the input end of the metabonomics verification module, candidate drug combinations are screened according to the cooperative scores, a sample after the prediction combination treatment is obtained, and visual display is carried out; the output end of the metabonomics verification module is connected with the input end of the feedback optimization and self-learning module, and metabonomics analysis is carried out on the candidate medicine combination to obtain a verification result; and the feedback optimization and self-learning module optimizes model parameters based on the verification result and feeds the optimized parameters back to the artificial intelligent prediction module. Preferably, the multi-source data acquisition and feature fusion module includes: Collecting disease gene expression, drug targets and structural information; Carrying out graph structure coding on the drug characteristics, and representing the target characteristics in an One-Hot or Word2Vec mode; the drug-disease-gene three-layer heterogeneous network is constructed and output in the form of an adjacency matrix or tensor. Preferably, the multi-source data acquisition and feature fusion module further comprises the step of jointly embedding the chemical structure of the drug and the target gene feature to form a drug multi-dimensional expression vector, wherein the formula is as follows: In the formula, Representing the vector for a multi-dimensional feature of the drug; representing a structural feature vector obtained from a pharmaceutical chemistry structure SMILES sequence via an embedded network; representing a target feature vector obtained from drug target gene/protein features via an embedded network; representing the operation of stitching two vectors in the feature dimension. Preferably, the artificial intelligent prediction module obtains an AI prediction result by adopting an AI prediction model with a multi-task learning structure, wherein a main task is prediction of a medicine pair cooperative score, and an auxiliary task is ta