Drug virtual screening method and device based on molecular similarity and semi-supervised learning

Abstract

The invention discloses a drug virtual screening method and device based on molecular similarity and semi-supervised learning. The method comprises the following steps: S1, collecting a data set to obtain a ligand molecule sample with a biological activity value and a ligand molecule sample without a biological activity value; s2, constructing a regression model by using the ligand molecule sample with the biological activity value obtained in the step S1; s3, calculating the similarity between molecules in the data set; s4, calculating triple loss by utilizing the molecular similarity obtained in the step S3 and the regression model obtained in the step S2; and S5, training the model according to the loss functions obtained in the steps S2 and S4. On the basis of a semi-supervised learning method, a large number of samples without experimental biological activity values are introduced into model training, model prediction values of the samples without the experimental biological activity values are constrained by using molecular similarity and triple loss, and the problem that a large number of samples without the biological activity values cannot effectively participate in model training in an actual application scene is solved.

Description

Technical field [0001] The present invention relates to a pharmaceutical virtual screening method and apparatus based on molecular similarity and semi-supervising learning, which can be used in the field of artificial intelligent drug design technology. Background technique [0002] Drug R & D has the characteristics of investment, high risk, and periodic length. In general, a drug research and development cycle is more than 10 years, R & D is invested in hundreds of millions of dollars, and presents a year-on-year trend. Drug-virtual screening is an important part of drug discovery, which can greatly reduce the time and cost of screening, which is of great significance for accelerated drug development. In recent years, with the development of artificial intelligent drug design, the machine learning algorithm has received a large number of applications in the pharmaceutical design, and the machine-learning-based drug virtual screening method has become the mainstream method, whic...

AI Technical Summary

Problems solved by technology

[0003] However, a good machine learning model often needs to use a large amount of ligand molecular data with known activity values ​​for training. However, in actual application scenarios, obtaining data sets with known biological activities through experiments is usually time-consuming and labor-intensive.

Images