Development of novel chemometric methods for constructing ADMET models

Novel chemometric methods termed partial least square coupled logistics regression (PLS-LR) and partial least square coupled cumulative logit regression (PLS-CLR) have been successfully developed and applied to overcome the problem of overfitting in quantitative structure-activity relationship (QSAR) modeling for absorption, distribution, metabolism, excretion and toxicity (ADMET) data.; In the first and second applications, based on novel chemometric methods, PLS-LR and PLS-CLR, categorical QSAR models for skin sensitization using Local Lymph Node Assay (LLNA) data have been constructed. In the third application, based on genetic function approximation (GFA), a methodology termed membrane-interaction QSAR (MI-QSAR) analysis has been used to develop QSAR models to predict drug permeability coefficients across cornea and its component layers (epithelium, stroma, and endothelium).; The QSAR models constructed not only are statistically significant, but have powerful predictive capability as well. This indicates that the novel chemometric methods developed and applied can handle the problem of overfitting.