Novel Molecular Modelling Methods Research For Protein-Protein/Ligand Interaction

Drug molecules play important roles in our life.As with the development,it is more and more difficult to develop a new drug with high activity,low resistance and low toxicity.Computer Aided Drug Design(CADD)can help us to save money and time efficiently.One of the most important thing in the CADD is to accurately evaluate the molecular interaction and binding free energy.It is important for us to study the interaction low and help to discovery the poteintial new drugs in short time with less cost.In this thesis,we used several theories,such as DFT-QSAR,free energy perturbation(FEP),molecular dynamic(MD)simulation,to develop different new computational method of molecular simulation.They were applied to study the molecular interactions including lead compound generation,drug resistance prediction,hotspots prediction on protein-protein interaction.Our method were proved to be effective and high accuracy after the validation.Last,several methods were developed to be web resources for helping non-expert researchers work easilyFirstly,we developed a new method to calculate Sterimol parameters based on DFT method.We constructed DFT-QSAR models based on Sterimol parameters and volumes,respectively.We compared two kinds of models by studying three systems.We found that the Sterimol-based model has better performance in desctibing steric effect than volume-based model.Among the three studied systems,the linear correlation coefficient is 0.89 for PDK1 series inhibitors.It has an almost 0.15 improvement than volume-based model.Then,we selected several molecules to perform an analysis which combined the binding mode and QSAR equation together.It finally proved the superiority of Sterimol parameters in describing steric effect of substituents than volume.Secondly,we developed in silico ligand directing evolution method to perform the hit to lead optimization.We constructed a molecule group library with 44 substituents.We replace hydrogen atoms on hit compound with substituents to generate potential lead compounds based on hit-receptor MD result and FEP theory.Because it just needs to perform MD simulation of hit-receptor complex,it may save more computational time.At the same time,we add a short time MD simulation at the refine step,thus will make the method to have a broader use.Moreover,the in silico ligand directing evolution method was validated by calculating 157 samples from 19 systems,which were collected from published papers.The accuracy on dividing samples into positive or negative was 93.6%.The linear coefficient R2 between calculation and experimental values was 0.82.The validation result proved the reliability of our method.We developed our method into web server named Auto in silico Ligand Directing Evolution(AILDE,http://chemvang.ccnu.edu.cn/ccb/server/AILDE/),which offered easy interface for medicinal chemists to use our method on drug molecule design.Thridly,we developed amino acid mutation scanning method to predict mutation-induced drug resistance based on FEP and MD simulation.It can simulate more mutation types and the short MD optimization on mutatns will make the bindig mode be close to real state.We used the method to predict 17 systems with 311 mutatns,which have experimental activity change on the published papers.Theses mutants have a broad distribution.We obtained a good agreement betwwen our calculation result and experimental value.It showed an accuracy 90%on dividing samples into resistant or non-resistant.Finally we constructed Auto in silico Macromolecule Mutation Scanning(AIMMS,http://chemyang.ccnu.edu.cn/ccb/server/AIMMS/)web server,which is the world's first de novo drug resistance prediction tool to predict multiple ligand-protein system.The easy input and multiple output formats of our web server will make it more convenient for more researchers to use.We use AIMMS to study ABA-PYR system and found 4 PYR mutants which could make ABA bind better which been proved by experiment.Lastly,we used MD simulation and FEP to study the protein-protein interaction(PPI)and developed a new protocol to detect interface residues and predict hot spot residues.First we preform the CAS on interface residues to predict the hotspot residues and mutation scanning were performed then on these hotspots for detecting more protein structure space.We collected 758 mutants with binding free energy change on PPI from 3 databases for method validation.The calculation result showed an accuracy 79.3%on qualitatively and R2=0.64 on quantitative evaluation.This result had a 10%improvement on R than other published best model.We constructed Protein Interface in silico Mutation Scanning(PIIMS,http://chemyang.ccnu.edu.cn/ccb/server/PIIMS/)web server,which is the world's first free web server for hotspots detection on protein-protein interface and the mutation scanning calculation on these hotspots,to help non-expert users to explore more protein structure space.