Prediction of inhibition of matrix metalloproteinase inhibitors based on the combination of Projection Pursuit Regression and Grid Search method

Quantitative structure–activity relationship (QSAR) models of three matrix metalloproteinases (MMP-1, MMP-9, MMP-13) inhibition were developed based on linear and non-linear modeling approaches by a series of N-hydroxy-α-phenylsulfonylacetamide derivatives (HPSAs). The Best Multi-Linear Regression (BMLR) method in CODESSA was used to select the most appropriate molecular descriptors from a large set of descriptors and develop a linear QSAR model. A combination of two methods, the Projection Pursuit Regression (PPR) and the Global Grid Search, was first used in generating the non-linear QSAR model of MMP inhibitory phenomena. Both the linear and non-linear modes could give good prediction results. Six models were built according to different MMPs and different MMPs inhibitory activities (log (106/IC50)). The prediction results of the PPR model were better than those obtained by the BMLR model, which proved that the non-linear model could better simulate the relationship between the structural descriptors and MMP inhibitory activities, even exploring MMPs active sites. It was proved that the combination of PPR and Global Grid Search method was a very useful modeling approach for the prediction of MMP inhibitory activities, and could be expected to be used in the other similar research fields.