Prediction Of Hot Regions In Protein-protein Interaction Based On Density Clustering And Vote Classifier

In the context of bioinformatics,the exploration of protein-protein interaction is becoming more and more mature.Nowadays,the prediction of hot spots and hot regions has become a key topic based on protein-protein interaction.Previous studies have shown that hot spots tend to be clustered,which is called hot regions.The research of hot regions prediction plays an important role in the exploration of protein function site and protein-protein interaction design.This thesis proposes a method named the prediction of hot regions in protein-protein interaction based on density clustering and vote classifier.First of all,according to the feature that the spatial density of hot regions is denser than that of non-hot regions,the density clustering algorithm is used to find some dense clusters to confirm the general locations of hot regions.Then,the vote classifier is applied to classify the residues in the predicted clusters.This vote classifier is composed of three kinds of base classifiers.These base classifiers,which called logistics regression,KNN classification algorithm and Robetta prediction model,extract the some attributs including the structural attributes,spatial distance attributes and energy attributes in predicted clusters respectively.The result of each based classifier is filtered by voting,which can enhance the stability of classification and remove some non-spots in predicted clusters effectively to improve the precision of predicted areas at the same time.Besides,on the basis of previous predicted results,the local community detection algorithm is employed to recycle related residues which are deleted by mistake before,which optimize the prediction precision further.Therefore,the optimized hot regions are the predicted hot regions in this thesis.The experimental results show that both the precision of the predicted hot regions and the precision of hot spot in hot regions improve a lot compared with previous model,which obtains better effects.