Prediction method of docking posture between protein and ligand based on graphic neural network

Abstract

This invention discloses a prediction method for docking poses between proteins and ligands based on graph neural networks. Firstly, a dataset of biological information samples of protein-ligand complexes is obtained, including sample data and annotated data. Secondly, docking pose generation models based on graph neural networks and multi-perspective docking pose evaluation models are constructed. The model parameters are further adjusted, and the structural generation model obtained through training is used to process the sample data, obtaining actual outputs of protein-ligand docking poses. Finally, mainstream docking structure evaluation metrics are used to assess the stability of the output results. This invention directly utilizes the biological structural information of ligands and proteins to generate optimal docking pose structures and evaluates the generated results through a comprehensive evaluation model from multiple perspectives. This improves the accuracy of docking pose structure prediction for ligand-protein interactions and enhances the effectiveness of evaluating the prediction results of ligand-protein docking pose structures.

Chuyue Liao

Junior Undergrad of Computer Science and Technology

My research interests include AI for Science, Deep Learning and Computer Vision.