Prediction Of Binding Affinity Between Protein-ligand Molecules Based On Graph Neural Network

The interaction between protein ligands often occurs in many basic biological activities.Understanding the interaction between protein ligands is of great significance for understanding many biological systems and assisting drug development.The way of quantitative study of interaction between protein and ligand is the determination of intermolecular binding affinity.Although traditional biochemical experiments or computational determination methods based on the first principles of physics are more accurate,obtaining the corresponding results requires high time or labor costs,and is not suitable for large-scale molecular screening in industry.The calculation method pursues a balance between practicality and affordability,allowing extensive exploration of the small molecule chemistry space.Therefore,it is necessary to study the calculation method of binding affinity between protein ligand molecules.In the past few decades,researchers have invested a lot of work in this field and proposed many calculation methods,but there are still some shortcomings.Among the many methods,methods based on deep learning stand out in terms of prediction accuracy.However,machine learning methods rely on manual feature engineering,and it is currently difficult to find a feature engineering method that is universal in the entire chemical small molecule space.Deep learning methods provide a new solution for this research through the end-to-end learning of complex neural networks.The prediction effects of these deep learning methods are higher than that of traditional scoring functions methods.But there is still room for improvement in the accuracy of prediction.In view of the emerging graph neural network has shown outstanding learning ability in other fields,its way of representing data in graph structure is very suitable for representing molecules and it has been less studied in the field of protein ligand binding affinity.In this article,the application of graph neural network in the prediction of intermolecular binding affinity of protein ligands is studied,and two prediction methods of intermolecular binding affinity of protein ligands based on graph neural network are proposed:1.Based on the structure of protein and ligand molecules,a cascaded graph convolutional neural network method called APMNet is proposed.This method regards molecules as graph data structures,and predicts the binding affinity between protein ligand molecules by cascading two different graph convolutional neural networks that can process non-European data.This method uses the respective advantages of the two graph convolutional neural networks to extract the features related to the binding affinity hidden in the molecular structure in stages,and use the features to predict the binding affinity.Experiments on the PDBBind v2016 data set show that APMNet has achieved good prediction results.Pearson R is 0.815 and RMSE is1.268.2.Aiming at a large number of protein molecules with known amino acid sequences but their structures have not been resolved,a method SGNet,which uses the structural information of the ligand molecular and the protein amino acid sequence information,is proposed.The graph attention neural network is used to learn the molecular representation of ligands,and the one-dimensional convolutional neural network is used to learn the molecular representation of the protein amino acid sequence initially encoded by CT(Conjoint Triad).The binding affinity between the two molecular pairs is predicted by fusing the characteristics of the two through the full junction layer.The generalization performance of SGNet model was investigated under three more rigorous cross validation strategies of new ligands,new proteins and new protein ligand pairs based on similarity clustering.The experimental results show that the predicted binding affinity of SGNet has little error compared with the real value,and has higher accuracy than the existing methods.Two methods are proposed to calculate the intermolecular binding affinity between protein and ligand,which aim at the protein with known structure and the protein amino acid sequence with unknown structure respectively.The research content of this paper can provide some guidance for the early stage of drug design and discovery.