Studies Of Quantitative Structure-Activity Relationship Of Bio-molecule Inhibitors Based On Three Different Algorithms

Quantitative Structure-Activity Relationship （QSAR） studies are importantresearch topics in chemoinformatics,and are an active area of the bioinformaticsresearch. They have been widely used for the prediction of various properties ofcompounds（physical and chemical properties, biologicalactivity） by the used ofdifferent statistical methods and various kinds of molecular descriptors. In this paper,different machine algorithms was adopted to build the QSAR model. And we focusedon the non-linear support vector machine method in the QSAR modeling.The first chapter of this paper reviews the study about QSAR, describes the basicprinciples and development of QSAR, detailed introduction the steps of thequantitative structure-activity relationship studies, and some QSAR modelingapproach.Cytosolic phospholipase A₂α, one of the three subtypes of Cytosolicphospholipase A₂（α, β and γ）, is deemed to play an important role in arachidonatepathway. Due to the rate-limiting provider for proinflammatory mediators, it is aparticularly attractive target for drug development. Studies have revealed that indolderivates compounds can inhibit the activities of Cytosolic phospholipase A₂α.However, few papers on the relationship between the molecular structure and theactivity of inhibitor was reported. In this study, the Quantitative Structure ActivityRelationship （QSAR） of indole derivates has been performed based on the dataset of49compounds. By using stepwise multiple linear regression,5descriptors wereselected from1777molecular descriptors, including GGI5（Topological charge indexG5）, TIE（dssC）（sum of E-State of atom type dssC:|2S（dssC））, RDF115a （the atomicSanderson ALOGP）, RDF100c （the atomic charge）, and RDF065p （the atomicpolarizability）. Subsequently, Partial Least Squares （PLS）, Artificial Neural Networks（ANN） and Support Vector Machine （SVM） were adopted to build the QSAR model,respectively. The independent test indicated the SVM can give best statistical results.And indole derivatives inhibitors actives might be related to global charge transfers,carbon atoms type linked benzyl sulfonamide and molecule geometrical the distancedistribution.We study of37c-Met kinase inhibitor the relationship between molecular structure and inhibitory activity based quantitative structure–activity relationship（QSAR） in this paper. A data set of37compounds was randomly separated into atraining set of30compounds, which was used to build the model and a test set of7compounds, which was applied to test or validate model. By using stepwise multiplelinear regression,6descriptors were selected from1666molecular descriptors, whichwere computed by E-DRAGON Software. Partial Least Squares （PLS）, ArtificialNeural Networks （ANN） and Support Vector Machine （SVM） were used to buildingQSAR models. The correlation coefficient R²for the test set were0.8204、0.8635、0.939responsibility; And the Root Mean Square Error （RMSE） for the test set were0.1640、0.1807、0.0033responsibility. The result in our’s work displayed that theSVM model had higher predictive capability to describing the relationship betweenthe structural descriptors and activity of c-Met kinase inhibitors compared with PLSand ANN.In our paper, Partial Least Squares （PLS） and Particle swarmoptimization-Support Vector Machine （PSO-SVM） were used to building QSARmodels for the prediction of inhibitory activity of a series of N-BenzoylatedPhenoxazines and Phenothiazines derivatives. QSAR model were validated by thesestatistical characteristics which were recommended by Tropsha. The result showedthat the PSO-SVM model had higher predictive capability for description of therelationship between the structural descriptors and activity of N-BenzoylatedPhenoxazines and Phenothiazines derivative compounds.