Abstract：The purpose of molecular virtual screening method is to find active molecules that can interact with a receptor protein and modify its biological behavior.The prerequisite of most molecular virtual screening methods is the known structure of protein or small molecule conjugate.However,for most proteins,this information is unknown.Therefore,this paper proposes a virtual screening method called Screener based on protein sequence alignment and active molecular similarity evaluation.Screener firstly generates the location-specific frequency matrix features,secondary structure features and solvent accessibility features from the sequence of receptor protein,then uses I-LBR program to predict the potential binding sites of receptor protein.Secondly,according to the predicted binding sites and related feature information,the template protein library is constructed.Then,all active molecules that interact with any template protein are collected to form a potential seed molecule library.Finally,the similarity between molecular 2D fingerprints is used to sort the set of molecules to be screened and complete the molecular virtual screening.On the benchmark datasets DUD40 and DUD-E65,Screener's average EF1% are 16.6 and 25.7,and HR1% are 44.1 and 67.6,respectively.The benchmark test results show that the average performance of Screener is better than that of AutoDock Vina,a virtual screening method based on docking,and FINDSITEfilt and PoLi,virtual screening methods based on structural alignment.
2022, Vol. 43 
