Pharmaceutical Perspectives of Nonlinear QSAR Strategies

The development of Quantitative structure-activity relationships (QSAR) models by machine learning methods as an attractive and helpful strategy in drug discovery is discussed. QSPRs approaches represent probably the most robust well known tools to mathematically analyze the correlation between molecular properties and the corresponding property of interest. Several key steps are involved in any QSAR modeling approach, data collection, calculation of suitable molecular descriptors from chemical structures and, if necessary, their selection, model generation, and finally, internal and external validation. In drug discovery, QSAR methodologies have demonstrated to be powerful in the prediction of simple chemical physical properties as well as complex pharmacodynamic, pharmacokinetic and toxicological profiles. Regarding the prediction of the toxicological profile of chemicals, in silico approaches are more and more applied as alternative methods to animal testing to refine and reduce animal experiments.