3results about How to "Reduce R&D Risk" patented technology

The invention discloses a molecular property prediction large language model construction method based on a knowledge graph. The method comprises a plurality of steps: firstly, acquiring pre-training corpora by using a high-throughput data acquisition technology, including related data such as molecular attributes and metabolic pathways; then, a knowledge graph about fourteen end points is constructed based on the data, and the knowledge graph represents the relation between the molecules and the attributes and the end points of the molecules in a triple form; thirdly, pre-training the large language model by using a LoRA technology, optimizing the model by adopting an incremental learning strategy, and forming an expert large model; then, a triple related to the target endpoint is retrieved from the knowledge graph through the RAG technology, and retrieval results are sorted; and filtering the retrieved triple through cue word design and an expert large model, and extracting key information. And finally, converting the filtered triple list into a natural language, inputting the natural language into an expert large model, generating a prediction result of molecular properties, and providing a reasoning basis. According to the method, the accuracy and efficiency of molecule and endpoint chemical reaction prediction are remarkably improved.

The invention discloses a plastic product research and development data processing method and system, and particularly relates to the technical field of plastic products. The extracted material performance complexity features and market feedback real-time features are comprehensively analyzed through fuzzy logic, so that the reliability of a product prediction model is accurately evaluated, according to the prediction result of the model, the result is further divided into an accuracy prediction result, a possible accuracy prediction result and an inaccuracy prediction result, and the prediction accuracy of the product is further improved. And corresponding treatment measures are taken according to different categories, so that the research and development efficiency is improved, the market adaptability and quality of products are ensured, enterprises can better meet the market requirements, and the overall competitiveness is improved.

This invention relates to a high-throughput prediction method and system for drug-regulatory element associations. The method includes: constructing a regulatory element-transcription factor network; constructing a regulatory element-target gene network; using the expression levels of transcription factors bound to the regulatory element and the target gene of the regulatory element as inputs, constructing a predictive model of the chromatin openness of the regulatory element; using gene expression levels under different drug action conditions as inputs, using the predictive model to predict and compare the openness of the regulatory element under different drug action conditions, thereby predicting drug-regulatory element associations. This invention can predict the relationship between regulatory elements and drugs at the genome-wide level with high throughput, enabling high-throughput screening of drug-regulatory element associations with relatively low investment, reducing drug development risks. These associations have potential value in research on drug mechanisms, drug combinations, and drug repositioning.