Research On Visual Virtual Screening System And Intelligent Toxicity Prediction Model For Small Molecules

The research and development of new drugs is a life science project with extremely high costs and risks and an extremely long cycle.Computer-aid Drug Design(CADD)has opened a new era of modern new drug research and development.However,there are still several problems that need to be resolved.First,facing the diverse and massive number of small molecules in the compound database,the integration of virtual screening is not satisfactory;second,it is impossible to find a more accurate and generalized toxicity prediction model in the prediction of molecular properties;third,facing Numerous virtual screening data and toxicity prediction data are unable to accurately capture key information and perform a comprehensive analysis of candidate molecules.Therefore,this thesis first analyzes the structure of different manifestations of molecules,and integrates two virtual screening methods.Secondly,this thesis applies deep learning methods to associate molecular structure with properties,and predict the toxicity of molecules through the rules of compound structure.Finally,this thesis applies information visualization technology to virtual screening based on pharmacophores to build a virtual screening integrated platform to achieve high-quality and high-performance virtual screening visualization.The details are as follows:(1)Visualized virtual screening analysis based on small molecules.This thesis combines advanced visualization technology to focus on the visualization of molecular three-dimensional structure and virtual screening data,including scientific computing visualization,screening data visualization,and molecular three-dimensional structure visualization.This paper uses Java Swing to support a variety of data models to realize the visualization of scientific computing;uses JFreeChart to draw bar graphs to realize the visualization of filtered data;uses Jmol to draw molecular ball and stick models,spherical models and cartoon models to achieve molecular three-dimensional Visualization of the structure.(2)Construct a molecular toxicity prediction model based on graph neural network.This thesis proposes an improved triplet information toxicity prediction model(TCTNet)based on graph neural network and combined with the chiral characteristics of molecules.This model uses local parity bits combined with an ordered set of neighbor nodes to capture local chiral features.TC-TNet also uses the attention mechanism module based on chiral triples to aggregate the triples to calculate the attention score.Finally,TC-TNet gathers neighbor information(including neighbor nodes and edges)according to the attention mechanism.Experiments have proved that the model has good operational capabilities and effectiveness.(3)Develop a virtual screening integrated system.In this paper,a virtual screening and visualization integrated system has been developed.Under the B/S architecture of visualization layer,logic layer and data access layer,the system integrates structure-based and ligand-based pharmacophore virtual screening methods,and uses Jmol to realize the three-dimensional structure display of microscopic small molecules.The system uses a variety of visualization methods to achieve good user interaction,including molecular editing,data management and analysis.The system solves the problem of low integration of virtual screening,reduces the coupling between functional modules,improves the accuracy of virtual screening,and speeds up operation and performance.