Studies of relationship between biological activities and HIV Reverse Transcriptase Inhibitors by Multivariate Adaptive Regression Splines with Curds and Whey

We developed a non-parametric modeling method specifically for the situation where the multivariate responses are correlated. The method is applied to a data set that contains the inhibitory activities against wild type HIV, the mutant strains 181 C, 103 N, 100 I and 188 L and the calculated Van der Waals and Coulomb interactions with the side chain and backbone parts of a selected set of 93 amino acid residues lining the binding pocket with 208 Non-Nucleoside RT inhibitors. It is shown that the model constructed can identify residues that are important to explain the relationship and presents clues to understand essential amino acids and physical interactions with residues.