Decoy Molecular Generation Algorithm Based On Physicochemical Properties And Structural Topology Parameters And Its Application

With the development of computer technology,information technology,structural biology,and theoretical chemistry,more virtual screening(VS)algorithms have been developed.How to evaluate these algorithms has become a research hotspot.The superiority of the VS algorithms is mainly embodied in the ability of distinguishing between molecules with specific target biological activity(Actives)and molecules without specific target biological activity(Inactives).Thus,the initial step in evaluating the algorithm is to prepare a benchmark data set,including active and inactive molecules for a specific target.The active data can be easily obtained through biological methods such as inhibition or agonist of enzymes,inhibition of cancer cells,and disease-related animal experiments.However,for specific test systems,active molecules can be easily gotten from scientific literature or some database,while inactive molecules,called trash molecules,are rarely reported.It's unbeneficial to construct the benchmark data set.Therefore,generating theoretical inactive molecules,namely decoy molecules,is particularly important for constructing the reasonable benchmark data set.The decoy molecule is defined as an theoretical inactive compound that is similar with the active compound in the physical and chemical properties,but dissimilar in the structure.Currently,DUD,DUD-E,and DecoyFinder are widely used virtual screening algorithm or database to generate decoy molecules.DUD contains 40 targets and the corresponding active and inactive compounds.But it has some shortcomings,such as,if the target is not contained in the original 40 targets,it couldn't generate decoy molecules,the amount of active compounds and decoy molecules for each target is not optimistic,and the Decoy molecules' skeleton diversity is low.DUD-E is an improved version of the DUD.DecoyFinder is a local executable software that can be used to generate decoy molecules.However,DUD-E and DecoyFinder still have a lot of deficiencies.First,the decoy molecules generated speed is slow.Second,DUD-E has a limited number of available targets(102 targets),even though a user can produce a decoy molecule,but an active molecule can produce no more than 50 decoy molecules for each active molecules,and the underlying database of DUD-E only rely on Zinc Database,lacking other database source,resulting in lack of compounds diversity.Moreover,some active molecules can not produce the corresponding decoy molecules in DUD-E.Although DecoyFinderproduces multiple decoy molecules for each active molecules,but the core algorithm of DecoyFinder does not take into account the effects of charge,resulting in low accuracy.Finally,with the increasing number of active molecules,how to choose a properate active query compounds for buliding the benchmark data set become difficult,and both DUD-E and DecoyFinder havn't taken it into account.In the present study,we developed an accurate,high-speed,large-scale decoy molecule generating program suite,term as RApid Decoy Retriever(RADER),which can quickly produce better decoy molecules to build benchmark data set to facilitate science research.The main contents of this study including:1)The first chapter summarizes the background of decoy molecule;2)The second chapter introduces RADER algorithm and its evaluation,and the achievement of RADER web;3)The third chapter introduces the application of RADER in PI3K-AKT-mTOR pathway;4)The fourth chapter summarizes and prospects the present study.This study has successfully developed a fast,accurate,parametric,easy-to-use program kit for generating decoy molecules and achieved its online version.RADER produces decoy molecules 7 to 550 times faster than DecoyFinder.We used MOE-docking and Autodock Vina enrichment rates to evaluate the quality of benchmark data set built by RADER.MOE-docking enrichment of RADER,DUD,DUD-E,and DecoyFinder,the average EFmax,EF1 and EF20 values were 35.2,36.1,36.4,and 37.0;8.5,10.4,11.7,and 10.7;and 2.6,2.5,3.2,and 2.8,respectivly.Autodock Vina enrichment of RADER,DUD,DUD-E,and DecoyFinder,the average EFmax,EF1 and EF20 values were 35.1,36.1,36.4,and 36.5;5.2,5.8,8.8,and 8;and 2,2.1,2.5,and 2.8,respectivly.The results show that the average enrichment rate of the RADER algorithm is smaller than that of DUD,DUD-E,and DecoyFinder,suggesting that Decoy molecules generated by RADER is more suitable for constructing the benchmark data set data set.In addition,this study successfully built the PI3K-AKT-mTOR pathway target benchmark data set using RADER.