Quantitative electronic structure-activity relationships of pyridinium cephalosporins using nonparametric regression methods

Projection pursuit regression (PPR) was applied to interpret and predict the antibacterial activity of pyridinium cephalosporins using semiempirical quantum mechanical descriptors. This method can deal with responses due to interactions of predictors (descriptors) which cannot be completely represented by additive regression models. Based on leave-one-out cross-validation, the best PPR model gave a cross-validated r(2) or q(2) value of 0.711, whereas the traditional method, multiple linear regression, and another additive nonparametric model, alternating conditional expectations, produced the best q(2) values: 0.233 and 0.324 respectively. Its ability to provide models with good predictive ability reveals that PPR is a valuable tool in quantitative structure-activity relationship studies.