Molecular Docking And Metabolite Classification Based On Multi-source Information Fusion

Protein-ligand docking and classification of active metabolites are two important issues in drug screening.However,the traditional structure sampling and scoring functions are not accurate enough for molecular docking.And the conventional mass spectrometry experiments cannot quickly classify the metabolites.Therefore,there is an urgent need to develop new protein-ligand molecular docking and metabolite classification methods.Here,we developed pmDock,a protein-ligand docking method based on multi-source information fusion.First,AutoDock and SwissDock were used to sample the structure.Then,the sampling results were screened by binding energy ranking.Next,we calculated the distances between the predicted ligands’ geometry center and removed the conformations if the distance is larger than 3 A.Then,we compare the distance between the predicted ligand geometry centers and experiments.Finally,the K-means algorithm is used to find the best conformation.The results show that(1)the prediction accuracy of the geometric center position of the molecular ligand has been improved(2)the prediction accuracy of complex binding sites and interactions has been improved.We also developed an active metabolites classification method based on multi-source information fusion.First,the VGG-16 network is used to extract the mass spectra features of metabolites.Then,hierarchical clustering and principal component analysis were carried out on the mass spectrum and physicochemical properties of the metabolites.Finally,PPV and STY values of this fusion method were calculated.The results show that(1)the classification of the number of nitrogen atoms,hydrogen atoms,and hydrogen donors better agree with the mass spectrum classification of metabolites;(2)these three physical and chemical properties can divide metabolites into two categories.The protein-ligand docking and metabolite classification method based on multi-source information fusion can improve the docking and classification results’ accuracy and provide new enlightenment on the related issues in future early drug screening.