Residue-Based Protein-Protein And Protein-Ligand Interactions

It has been demonstrated that while docking programs can generating poses with acceptable geometries close to the crystal structure successfully,virtual screening cannot always produce acceptable hit rates.Empirical scoring functions are typically derived from a collection of protein-ligand complexes.It is hard to fit a universal scoring function that meets all types of targets.Furthermore,in general scoring functions the noncovalent protein-ligand interactions are normally treated as additive which is not always applicable.For virtual screening,a major issue is to improve the ranking power to discriminate actives and inactives.In structure-based virtual screening,removing false positives of docking results is one of the most challenging works.Tailored strategies beyond the application of a single score list of standard scoring function are recommended for virtual screening applications.We describe a classification model based on detailed protein-ligand interaction decomposition and machine learning.Protein-ligand empirical interaction components(PLEIC)are used as descriptors and support vector machine learning is used to develop a target specific classification model(PLEIC-SVM)to discriminate actives and inactive molecules.Experimentally derived activity information is used for the model training.An extensive benchmark study on 36 diverse data sets from the DUD-E database has been performed to evaluate the performance of the method.The results show that the PLEIC-SVM model performs much better than standard empirical scoring functions in ranking power in structure-based virtual screening.The trained PLEIC-SVM model is nonadditive which is helpful to account for the coupling interactions and the model is able to capture important interaction patterns between ligand and protein residues for one specific target.The theoretical calculation of protein-protein binding free energy is a challenging issue in computational biology.It has been demonstrated that although protein-protein interfaces usually comprehend a large number of residues,only a few residues in the protein-protein interface contribute to the binding free energy:hot spots.PPIs disorders can cause many diseases,mainly due to the abnormal changes including lacking of salt bridge,weakening of hydrophobic interaction,the formation of steric hindrance.The variation of hotspot areas is one of the most important factor causing PPI disorder.Accurate prediction the specific and quantitative contributions of critical residues to protein-protein binding free energy is extremely helpful to reveal binding mechanisms.MM/PBSA method has been successfully used to predict hot-spots in protein-protein interface.In these studies,the entropy changes of the wild type and its mutant complexes upon binding are usually assumed to be canceled and are neglected in most calculations.Our research is to study the influence of entropy to hot-spots prediction systematically.We propose an interaction entropy approach combined with the MM/GBSA method to compute residue-specific protein-protein binding free energy.In this approach,the entropic contribution to binding free energy of individual residues uses the interaction entropy method which is explicitly computed from MD simulation.The interaction entropy approach calculating the entropic contribution to binding free energy is determined from fluctuation of the gas-phase interaction energy in MD simulation.An extensive set of realistic protein-protein interaction systems are studied.The results show that the computed residue-specific binding free energies are in better agreement with the corresponding experimental data by including the entropic contribution.