Hot Spots Prediction Algorithm For Protein-Protein Interaction Using F-score And MRMR

It is the main topic that prediction of hot spots at protein interfaces in functional proteomics.Clackson found that most of the binding energies in hGH-hGHbp interactions were contributedby a few significant residues which were called hot spots at the protein-protein interfaces. Thesesignificant residues are not uniformly distributed along the interface; rather they are clusteredwithin locally tightly packed regions, around the hot spots we observe moderately conservedresidues.Considering the protein sequence, structure and the interface，prediction of hot spots is acomplicated problem. Up to now, the biological properties that are responsible for hot spots havenot been fully understood. Although the existing methods for predicting hot spots had madesome achievements, these methods still had to improve. These methods combined with machinelearning algorithms avoid the shortcomings of the experimental period, computationally costlyand small-scale application. This thesis presents a method for prediction of hot spots based on avariety of characteristics. Firstly, extract a variety of related protein sequences and structuralcharacteristics; Secondly, use F-score and mRMR to get protein amino acid physicochemicalfeatures (such as the weighted hydrophobic and the weighted residue contact) and structuralfeatures (such as solvent accessible surface area and protrusion index) to predict the hot spotsusing SVM.In the prediction of hot spot residues, the proposed method has achieved the expected goaland it was proved to be feasible compared with other methods. With the data extracted fromASEdb and BID respectively, the experimental results show high accuracy. We define hotregions based on community structure in complex network by preliminary work. And how torealize the proposed algorithm is the following research work.