CN-121983350-A - Medicine curative effect and toxicity intelligent evaluation method and system based on multi-modal pathology data fusion

CN121983350ACN 121983350 ACN121983350 ACN 121983350ACN-121983350-A

Abstract

The invention discloses a drug efficacy and toxicity intelligent evaluation method and system based on multi-modal pathology data fusion, and relates to the technical field of drug evaluation systems. The method comprises the steps of collecting biological characteristic data under the action of a target drug, and constructing multi-mode characteristic vectors, wherein the biological characteristic data comprises animal experiment data sources, in-vitro experiment data sources and public databases or existing research data. The invention carries out quantitative analysis on the change of the multi-mode characteristics in the time dimension, fuses the characteristics reflecting focus improvement, immune activation and curative effect stability to construct curative effect indexes, simultaneously carries out weighted integration on the characteristics reflecting tissue injury, immune imbalance and toxicity related effects to construct toxicity indexes, and on the basis, realizes multi-dimensional comprehensive judgment on the effect of the medicament by establishing a curative effect and toxicity combined evaluation system and divides the quality grades of different medicament schemes in an intuitive mode.

Inventors

XU JING
DENG XUETING
YU YUYANG
HE JINLI
HU XUEMEI
WANG RONG
SUN KAI
YANG MENGXI
XIONG QUANXIN
ZHONG ZHENDONG
HE QINGHUA
LUO DAN
ZHAO YANYING
TONG YAN
Lin Yaqiu
WANG MEILIN

Assignees

四川里来思诺生物科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (8)

1. A drug efficacy and toxicity intelligent evaluation method based on multi-modal pathology data fusion is characterized by comprising the following steps: Collecting biological characteristic data under the action of a target drug, and constructing a multi-modal characteristic vector, wherein the biological characteristic data comprises an animal experiment data source, an in-vitro experiment data source and a public database or existing research data, and the multi-modal characteristic vector comprises a pathological phenotype characteristic vector, an immune response characteristic vector, a drug resistance characteristic vector and an effect characteristic vector; taking the acquired multi-modal feature vector as input, establishing a multi-layer nonlinear mapping structure, mapping the target biological feature to a human feature space, and outputting a human feature prediction result; correcting the human body prediction feature vector by applying a physiological constraint rule; Constructing a drug efficacy index and a toxicity index based on the corrected predicted feature vector, and calculating an efficacy-toxicity comprehensive score; And generating a drug development decision report according to the comprehensive score.
2. The intelligent evaluation method for the curative effect and the toxicity of the medicine based on the multi-modal pathology data fusion according to claim 1, wherein the construction method for the pathological phenotype characteristic vector comprises the following steps: Dividing the single drug types in the drug set into a plurality of functional subclasses according to the functions of the drugs; Based on the functional subclasses, extracting corresponding biological characteristic data of the single medicaments in each subclass under each time node respectively, and carrying out weighted fusion on the biological characteristic data at the same moment in the same functional subclass, so as to construct a subclass-time dimension intermediate characteristic matrix; summarizing the characteristics of each function subclass at each time node to form a characteristic vector of the function subclass at each time node; And constructing pathological phenotype feature vectors corresponding to the same time node by integrating feature vectors of all the functional subclasses at the same time node.
3. The intelligent evaluation method for the curative effect and the toxicity of the medicine based on the multi-modal pathology data fusion according to claim 1, wherein the construction method of the immune response characteristic vector comprises the following steps: Extracting immune feature vectors of each time node from the immune index data set, and time-aligning the immune feature vectors with pathological phenotype feature vectors for enabling immune response of each time node to correspond to the pathological phenotype; grouping the immune indexes according to the functional subclasses, and constructing functional subclass-time node immune characteristics which are used for guaranteeing that the characteristics are consistent with pathological phenotype characteristics in dimension; weighting and fusing the immune characteristics of the same functional subclass at a time node to form a functional subclass fusion vector; And calculating time weight according to each time node, and weighting the function subclass fusion vector according to time to obtain a final immune response characteristic vector.
4. The intelligent evaluation method for the curative effect and the toxicity of the medicine based on the multi-modal pathology data fusion of claim 1, wherein the construction method for the drug resistance eigenvector comprises the following steps: Aligning the pathological phenotype feature vector and the immune response feature vector based on the time node, thereby forming a joint feature vector; acquiring characteristic variation of the joint characteristic vector under adjacent time nodes; Constructing a drug resistance index according to the characteristic variation; And the drug resistance indexes of all the time nodes are connected in series to form a drug resistance characteristic vector.
5. The intelligent evaluation method for the curative effect and the toxicity of the medicine based on the multi-modal pathology data fusion according to claim 1, wherein the construction method for the effect characteristic vector comprises the following steps: Extracting tissue characteristic data from the pathological phenotype characteristic vector, wherein the tissue characteristic data is used for describing the structure and state of biological tissues where the medicine acts; Extracting immune cell distribution characteristic data from the immune response characteristic vector, wherein the immune cell distribution characteristic data is used for representing infiltration density, spatial distribution mode and dynamic migration trend of various immune cells in a drug action micro-domain; Extracting time evolution characteristic data from the drug resistance characteristic vector, wherein the time evolution characteristic data is used for tracking the attenuation rule and the adaptive change of the curative effect response in the drug action micro-domain along with the time; Performing dimension alignment and standardization processing on the extracted tissue characteristic data, immune cell distribution characteristic data and time evolution characteristic data, and eliminating dimension differences and distribution deviations among different characteristics; and fusing the standardized tissue characteristic data, the immune cell distribution characteristic data and the time evolution characteristic data to form an effect characteristic vector.
6. The intelligent evaluation method for the curative effect and the toxicity of the medicine based on the multi-modal pathology data fusion of claim 1, wherein the construction method of the multi-layer nonlinear mapping structure comprises the following steps: Collecting associated data corresponding to cross-species multi-modal characteristics of a target organism and a human body, and constructing a cross-species characteristic mapping data set according to the associated data; performing homology alignment and feature space standardization processing on the cross-species feature mapping data set; and constructing and outputting a multi-layer nonlinear mapping structure for realizing the mapping of the target biological characteristics to the human body characteristic space based on the processed cross-species characteristic data.
7. The method for intelligently assessing the efficacy and toxicity of a drug based on multimodal pathology data fusion according to claim 1, wherein the method for efficacy-toxicity comprehensive scoring comprises the following steps: According to the human body characteristic prediction result, the therapeutic effect index and the toxicity index of the medicine are respectively obtained; based on the efficacy and toxicity index, efficacy-toxicity balance evaluation is carried out on the human body characteristic prediction result obtained in real time, so that comprehensive evaluation data of drug efficacy and toxicity are generated.
8. A drug efficacy and toxicity intelligent assessment system based on multi-modal pathology data fusion, using the drug efficacy and toxicity intelligent assessment method based on multi-modal pathology data fusion of any one of claims 1-7, comprising: The data acquisition and preprocessing module is used for acquiring biological characteristic data of the target medicine set from the multi-source heterogeneous data source and preprocessing the acquired data; the multi-mode feature construction module is used for constructing pathological phenotype feature vectors, immune response feature vectors, drug resistance feature vectors and effect micro-domain feature vectors, and carrying out functional subclassification, time sequence integration, weighted fusion and attention mechanism weighting treatment; the cross-species mapping module is used for mapping the target biological feature vector to the human body feature space to realize cross-species prediction; the physiological constraint correction module is used for carrying out physiological constraint correction on the human body prediction characteristic vector so as to ensure reasonable prediction results; the comprehensive evaluation module is used for constructing curative effect indexes and toxicity indexes based on the corrected human body characteristic prediction results, and generating curative effect-toxicity comprehensive scores and medicament comprehensive evaluation reports.

Description

Medicine curative effect and toxicity intelligent evaluation method and system based on multi-modal pathology data fusion Technical Field The invention relates to the technical field of drug evaluation systems, in particular to a drug efficacy and toxicity intelligent evaluation method and system based on multi-mode pathology data fusion. Background The medicine evaluating method based on the multi-mode pathological data fusion is a research method for integrating multi-source heterogeneous data such as genomics, imaging, biomarkers, clinical texts and the like based on large model technology so as to systematically evaluate the curative effect and toxicity of the medicine. Through searching and the Chinese patent application of the publication number CN120199516A, a drug effect prediction method for drug development is provided, the interaction strength between a drug and a target point is predicted through a model, and comprehensive analysis is carried out by combining with a disease background, so that potential drug candidate molecules can be more accurately identified, and the development period is further shortened. However, in the practical use process, the above disclosed method and similar prior art schemes are mostly based on static features or single time node data for analysis, lack of systematic characterization of dynamic changes of multi-modal features in the time dimension, and are difficult to accurately reflect the evolution trend of key factors such as focus change, immune response and efficacy stability in the drug action process, meanwhile, for toxicity evaluation, the prior art generally depends on only one or a few toxicity indexes, and fails to comprehensively analyze in combination with tissue damage, immune imbalance and various toxicity related effects, so that toxicity evaluation results are unilateral. Disclosure of Invention The invention aims to provide a drug efficacy and toxicity intelligent evaluation method and system based on multi-modal pathology data fusion, so as to solve the problems in the background technology. In order to achieve the above purpose, the present invention provides the following technical solutions: in a first aspect, a method for intelligently evaluating efficacy and toxicity of a drug based on multi-modal pathology data fusion is provided, including: Collecting biological characteristic data under the action of a target drug, and constructing a multi-modal characteristic vector, wherein the biological characteristic data comprises an animal experiment data source, an in-vitro experiment data source and a public database or existing research data, and the multi-modal characteristic vector comprises a pathological phenotype characteristic vector, an immune response characteristic vector, a drug resistance characteristic vector and an effect characteristic vector; taking the acquired multi-modal feature vector as input, establishing a multi-layer nonlinear mapping structure, mapping the target biological feature to a human feature space, and outputting a human feature prediction result; correcting the human body predictive feature vector by applying a physiological constraint rule so as to ensure the physiological rationality of the predictive feature; Constructing a drug efficacy index and a toxicity index based on the corrected predicted feature vector, and calculating an efficacy-toxicity comprehensive score; And generating a drug development decision report according to the comprehensive score. As a further preferable mode of the technical scheme, the construction method of the pathological phenotype characteristic vector comprises the following steps: Dividing the single drug types in the drug set into a plurality of functional subclasses according to the functions of the drugs; Based on the functional subclasses, extracting corresponding biological characteristic data of the single medicaments in each subclass under each time node respectively, and carrying out weighted fusion on the biological characteristic data at the same moment in the same functional subclass, so as to construct a subclass-time dimension intermediate characteristic matrix; summarizing the characteristics of each function subclass at each time node to form a characteristic vector of the function subclass at each time node; And constructing pathological phenotype feature vectors corresponding to the same time node by integrating feature vectors of all the functional subclasses at the same time node. As a further preferred embodiment of the present invention, the method for constructing the immune response feature vector includes: Extracting immune feature vectors of each time node from the immune index data set, and time-aligning the immune feature vectors with pathological phenotype feature vectors for enabling immune response of each time node to correspond to the pathological phenotype; grouping the immune indexes according to the functional subclasses, and constructing fun