Drug recommendation system for regulating and controlling disease targets based on three-channel deep learning, computer equipment and storage medium

Abstract

The invention discloses a drug recommendation system for regulating and controlling disease targets based on three-channel deep learning, and belongs to the technical field of drug relocation and three-channel deep learning model prediction. The system comprises a three-channel deep neural network model, and the three-channel deep neural network model comprises a sequence processing model composed of a feature extraction network, a three-channel feature optimization network and a full connection network. The three channels are respectively a positive sample drug molecule SMILES sequence, a negative sample drug molecule SMILES sequence and a target amino acid sequence, and the sequences are input into a feature extraction network for sufficient feature extraction; the features are input into a three-channel feature optimization network, the distance between a negative sample drug molecule and a target point is maximized, the distance between a positive sample drug molecule and the target point is minimized, the learned feature representation is optimized, and the three optimized feature vectors are spliced together to form a spliced vector; and the splicing vector and the drug-target binding affinity value are input into a fully connected network for regression analysis, and finally a predicted value of the drug-target binding affinity is output.

Description

technical field [0001] The invention relates to the technical field of drug-target positive and negative sample division and three-channel deep learning model prediction, in particular to a drug recommendation system, computer equipment, and storage media based on three-channel deep learning to regulate disease targets. Background technique [0002] At present, there are more than 4,500 known diseases, and 90% of the diseases have no curative drugs. In the face of public health emergencies, the lack of effective drugs threatens the lives of the people and brings huge economic losses to the country. Currently, there are few systems for recommending effective drugs for disease targets through drug-target binding affinity prediction. [0003] The identification of drug-target interactions is the key to drug repositioning, which can be divided into two categories: predicting drug-target binding affinity (DTA) using regression methods, and predicting the probability of drug-targe...

AI Technical Summary

Problems solved by technology

In DTI, negative sample information needs to be considered. At present, the division of positive and negative samples is mainly based on the results of chemical experiments, and the Davis data set uses dissociation constants to measure the binding affinity of the drug-target. The digital standard for dividing positive and negative samples is not clear.

Images