Research On Cavity Dynamics Visual Analysis And Prediction Of Allosteric Pockets Based On NMA

Cavity is an important part of the protein molecule and has a significant impact on the regulation of protein function.There are multiple cavities in proteins,and the cavities including active sites are key structures in the regulation of protein function.However,the constant movement of cavities with the protein structure makes it difficult for domain experts to observe and understand them,so visualisation techniques to understand cavities have become one of the important topics in the field of bioinformatics.The regulation of protein function be achieved by intervening in the allostery motion of pockets(a special case of cavities).However,the identification of allosteric pockets by biological experiments is time-consuming and expensive,so an approach to allosteric pockets by means of machine learning techniques has emerged.In this paper,we investigate the dynamic visual analysis of cavities and the prediction of allosteric pockets.Firstly,a new cavity dynamic visualisation approach is proposed for the interactive exploration of protein molecular cavities and all amino acids to address the lack of noncavity-lining amino acids in the cavity dynamic visualisation task.The method uses the dynamical information provided by the normal mode analysis to design a visualisation scheme to characterise the cavity and amino acids dynamics,facilitating the user to analyse the cavity dynamics from the direct and indirect perspectives.The user obtains correlations between amino acids and cavity motions through interactive exploration of visual analysis system,providing a new perspective for allosteric pockets identification.Secondly,to improve allosteric pockets prediction,this paper proposes a new allosteric pockets prediction method consisting of extreme gradient enhancement and conformational perturbation.The model performs pocket detection and extracts pocket features on the protein surface,and uses an XGBoost-based method to construct a classification model.An allosteric pockets identification method based on probe perturbation is used to examine the effect of probe molecules on changes in protein dynamics to discover allosteric pockets.Finally,the two models are combined to construct a more explanatory and accurate method.Finally,the visual analysis system and the allosteric pockets prediction method are applied to biological field cases and public datasets for analysis to validate the effectiveness and practicality of the proposed method.