Research On Complementarity Of 2D And 3D Molecular Similarity Virtual Screening Methods And Their Applications

The drug discovery process often requires a long period of time.The chances of finding large quantities of drugs from a large number of natural products and synthetic compounds are very small.Drug development is a very time-consuming,labor-intensive and high-risk project.With the continuous development of hardware and technology associated with drug discovery,more and more compound structures have been resolved and analyzed.The law of compound druggability has been concluded,and it is easy to carry out high-throughput screening or virtual screening(VS)of millions or even tens of millions of compounds using computer.Currently,computer-aided drug design(CADD)is gradually becoming one of the most important tools for drug discovery due to its low-cost,high-efficiency,and customizable features.Among them,the ligand-based VS methods have become the mainstream technologies in the lead compound discovery stage because of its unnecessity to analyze the crystal structure of protein target.It mainly includes 2D molecular fingerprints(based on structural fragment information)and 3D features calculation of efficacies(shape,electrostatic potential,pharmacophore etc.).2D fingerprints-based VS tools uses the binary structure fragment information between compounds,which have the characteristics of fast speed and high hit rate.However,limited by the predefined fragment structure,it is difficult to find novel scaffolds in the VS process.VS tools based on 3D molecular shape(electrostatic potential and pharmacophore)takes more time to calculate,and the accuracy is generally lower than that of 2D fingerprint methods,but it will not be limited by chemical fragments,and the active compounds identified by 3D similarity methods often have novel scaffolds.The advantages and disadvantages of 2D and 3D molecular similarity VS methods are like the two semicircles of the Taiji diagram.Its core(principle)is based on the two-dimensional(2D)and three-dimensional(3D)characteristics of the molecule.Will it be possible to hybridize them to get a complete "round" ? On the basis of the research on the hybrid molecular similarity calculation in our group,in this study we systematically explored the general laws of the complementarity of 2D and 3D molecular similarity methods.To achieve this,two benchmark data sets,DUD-E(102 targets)and LIT-PCBA(15 targets)were employed to evaluate the systematical efficiency of 15 3D molecular similarity-and 11 2D similarity-based VS tools.The evaluation indicators include(1)“screening power”,such as the area under the ROC curve(Area Under Curve,AUC),enrichment factor(Enrich Factor,EF),hit rate(Hit Rate,HR);(2)“scaffold hopping power”,such as analysis of the number of scaffolds and overall similarity analysis of the scaffolds;(3)complementarity analysis,such as templates complementarity analysis and methods complementarity analysis.Finally,we rescored the 2D and 3D methods using a predefined scoring function,and systemically comparison VS performance of the hybrid similarity(Hybrid Sim),pure 2D and 3D similarity tools.The results showed that:(1)among the 15 selected 3D similarity VS tools,SHAFTS,LS-align,Phase Shape＿Pharm and LIGSIFT showed better VS accuracy,and also exhibited better scaffold hopping ability;(2)compared with 3D similarity methods,most of 2D similarity methods show better VS performance;(3)compared with the pure 2D and 3D similarity VS methods,Hybrid Sim can effectively improve VS performance through a predefined scoring function.Based on the optimal hybrid similarity combination methods,we developed an online virtual screening platform,which integrates a pre-processed compound database,four optimal 3D molecular similarity methods and two optimal 2D similarity methods.We truly believe that this platform will be helpful for drug discovery.