Using Ligand Similarity Profile For Drug Design

Currently,pharmaceutical industry is falling into the mire of “high investment but low productivity”.Quantitative structure-activity relationships(QSAR)can establish relationship between chemical structure and its biological activities.The application of QSAR can guide the optimization of compound and improve the success rate of drug discovery.Effectively characterize the molecular information is a key problem during the construction of QSAR models.In this thesis,we proposed two molecular descriptors by calculating the ligand similarity profile between objective compounds and diverse biological ligands.Then,we explored the application of these methods in several QSAR studies.First,we introduced a molecular descriptor,named biologically relevant spectrum(BRS).The BRS of an objective compound is its two-dimensional similarity profile with 2000 diverse endogenous compounds.We developed several QSAR models to explore the capacity of BRS in the prediction of ADME/T properties.(1)We constructed five acute toxicity and two aquatic toxicity prediction models.In acute toxicity models,10-fold cross-validation squared correlation coefficient(Q2,for training sets)and determination coefficient(R2,for test sets)were in the range of 0.543~0.655 and 0.566~0.840,respectively.In fathead minnow and tetrahymena pyriformis toxicity discrimination models,the overall prediction accuracies(Qtotal)for test sets were more than 90% and Matthews correlation coefficient(MCC)were more than 0.7.(2)We developed several prediction models for blood-brain barrier(BBB)penetration,human intestinal absorption(HIA)and P-glycoprotein(P-gp)inhibitors.In the BBB regression model,R2 and RMSE for test set were 0.762 and 0.41.In the HIA discrimination model,Qtotal and MCC were 98% and 0.789 for test set,and the model could correctly predict 93.8% compounds of 634 oral drugs.In the P-gp inhibitors discrimination model,Qtotal and MCC were 78.1% and 0.537 for test set.The results showed that BRS performed similar or superior to traditional molecular descriptors in the prediction of ADME/T properties.Then,we improved BRS to consider 3D information of ligands.The 3D similarity profile that calculated between objective compounds and BRCD-3D was named three-dimensional biological relevance spectrum(BRS-3D).We applied BRS-3D in the prediction of ligands selectivity for several important therapeutic targets.(1)We explored the selectivity prediction of estrogen receptor(ER)ligands.The bioactivity data were extracted from ChEMBL.Four types of prediction models were constructed,i.e.,pair-wise selectivity regression model,activity regression models,pair-wise discrimination model and functional discrimination model.Most of the models were performed well.The results showed that BRS-3D performed similarly to classical 2D descriptors and better than classical 3D descriptors.Combining BRS-3D and 2D descriptors can further improve the prediction performance.(2)We investigated the ligand selectivity of two important GPCRs,cannabinoid receptor(CB)and dopamine receptor(DR).In the CB selectivity study,the models based on Ki and IC50 activity data provided good prediction ability.Comparison of different descriptors,BRS-3D performed better than SubFP and MACCS fingerprint,and similar to 3D MoRSE and GETAWAY descriptor.In the study of DR selectivity,we constructed five subtype selectivity regression models between DR subtypes.Q2 and R2 of the models were in the range of 0.578~0.680 and 0.602~0.753,respectively.Then,four pairwise and a multi-type(D2,D3,D4)classification models were developed with the prediction accuracies around or over 90%(for test sets).Furthermore,we found the important ligands in BRCD-3D that can differentiate selective ligands,which could provided a guidance for the study of ligand selectivity.In conclusion,the ligand similarity profile based method could effectively describe molecules,and could construct predictive ADME/T and selectivity models.The 2D method(BRS)has the advantages of rapid and accurate,but unable to achieve scaffold hopping.BRS-3D get rid of the limit of 2D topological structure,and can take various conformations into account.This advantage makes BRS-3D well applied for the prediction of conformation related endpoints.But the calculation of BRS-3D is time-consuming,and BRS-3D could not describe the pharmacophore features well.In the further drug design,we can merge the 2D,3D and pharmacophore features to improve the efficiency of virtual screening.