Evaluation,Optimization And Application Of Molecular Docking Based Virtual Screening Methods

Hit identification is one of the key steps in drug discovery and development.The emergency of new technologies,such as combination chemistry and high-throughput screening(HTS),has significantly accelerated the synthesis and screening speed of new compounds,but their high cost and low success rate still severly limit their applications in hit discovery.As a new technology complementary to HTS,virtual screening(VS),has attracted more and more attention from research institutions and pharmaceutical companies.VS can effectively enrich potentially active molecules from a large number of virtual compound libraries for specific targets,thereby reducing the number of tested compouds in the experimental screening stage and improving the success rate of screening.In parallel with the deteminiation of plentiful three-dimensional structures of biomolecules,the past few years have witnessed the obvious advantages of docking based virtual screening(DBVS),and now it has been rountinly used in hit discovery.However,the results of DBVS in actual use vary widely due to the imperfect molecular docking theory and numerous docking programs with different performance.Therefore,how to improve success rate by customizing and optimizing the screening scheme is an important research topic in DBVS.In this thesis,we first systematically evaluated the performance of mainstream docking programs on different types of targes,and then proposed several improved methods and strategies to overcome the deficiencies of current docking programs.Finally,our methods were applied to a real-world VS campaign and desired outcomes were achieved.The main contents and results of this thesis are as follows:(1)The performance of sampling power and scoring power of 10 popular docking programs were systematically evaluated,including AutoDock,AutoDock Vian,LeDock,rDock,UCSF DOCK,LigandFit,Glide,GOLD,MOE Dock,and Surflex-Dock.Among these programs.GOLD and LeDock had the best sampling power(GOLD:59.8%accuracy for the top scored poses;LeDock:80.8%accuracy for the best poses).AutoDock Vina possessed the best scoring power(the Pearson's correlation coefficient rp and the Spearman's rank coefficient rs between the scoring value and the experimental value of the top socre poses are 0.564 and 0.568,respectively).Overall.the ligand binding poses could be identified in most cases while the ranks of the binding affinities to the entire dataset could not be well predicted by most docking programs.However,to some types of protein families,relatively high linear correlations between the docking scores and experimental binding affinities could be achieved(2)Based on the 2P2IDB database,we explored the structural features of the known small-molecule PPI inhibitors and analyzed the characteristics of the PPI binding pockets.respectively.Our results indicate that the chlorinated conjugate group and amide-like linkage are two types of privileged fragments of PPI inhibitors;the average druggability of the binding sites of the PPI targets in 2P2IDB is slightly worse than that of traditional ones.More importantly,we evaluated the sampling power and screening power of six popular docking programs for PPI targets.The tested docking programs exhibit an acceptable accuracy on pose prediction for PPI inhibitors,but their screening powers for identifying PPI inhibitors are still not satisfactory(3)We constructed a non-redundant pose ranking benchmark dataset for small-molecule PPI inhibitors.Then,the performance of two Prime MM/GBSA procedures from the Schrodinger suite and seven different MM/PB(GB)SA procedures from the Amber package were evaluated to identify the correct binding poses for PPI inhibitors Our results sho wed that MM/PBSA outperformed the Glide SP scoring function(success rate of 58.6%)and MM/GBSA in most cases,especially the PB3 procedure which can achieve an overall success rate about 74%.Moreover,the GB6 procedure(success rate of 68.9%)performs much better than the other MM/GBSA procedures.In addition,we developed the web server of Fast Amber Rescoring for PPI Inhibitors(http://cadd.zju.edu.cn/farppi/),which offers a freely available service for rescoring the docking poses for PPI inhibitors.(4)The binding affinities of a series of pyrimidinone based Ubiquitin-specific protease 7(USP7)inhibitors were predicted by molecular docking,MM/PB(GB)SA,and umbrella sampling(US)methods.The results showed that flexible docking could not enhance the accuracy of binding affinities prediction,whilte MM/PB(GB)SA methods could obtain improved results.GB1 procedure has the best overall performance(the average of squared Pearson's correlation coefficient R2 between predicted and experimental is 0.412).The potential of mean force(PMF)for the dissociation process of inhibitors from USP7 binding pocket could accurately describe the binding affinities of inhibitors(R2=0.896).In addition,according to the US trajectories,we also compared and analyzed the unbinding pathway of four representive inhibitors.Finally,significant differences in the dissociation pathways between strong binding inhibiots and weak binding inhibitors were found.(5)In compliance with the strategy of "correct binding pose first",a VS towards the non-catalytic site of USP7 was conducted by combining molecular docking,MM/GBSA binding free energy calculation,and 3D shape based similarity searching methods in a hierarchical and feedback mode.A total of nearly 200 compounds were purchased and tested,and finlly several hits with novel scaffolds and good activities were discovered.