(506f) Developing New Models to Predict Drug Properties

Traditional drug design research is time intensive and costly; as such, efforts to model important biological properties in silico offer an effective alternative. These models allow researchers to screen efficiently more potential drugs, with significantly less experimentation. Current property models, however, have significant limitations, including high prediction errors, reliance on linear models, use of general-purpose heuristic algorithms for molecular screening, and applicability of the property models to limited chemical classes.

To address these problems, we have sought to develop improved algorithms for non-linear, quantitative structure-property relationship (QSPR) models based upon representative molecular properties. To supply these properties to the drug design models, we have developed new structure?based property models for aqueous solubility, octanol?water partition coefficient, skin permeation, infinite-dilution activity coefficient, melting point, critical volume and cytotoxicity. In general, the quality of our QSPR model predictions for these properties is comparable to, or better than, the present literature models. More importantly they form a valuable basis for our current efforts to design improved chemical penetration enhancers.