Comparative study of classification algorithms using molecular descriptors in toxicological databases

Abstract

The rational development of new drugs is a complex and expensive process, comprising several steps. Typically, it starts by screening databases of small organic molecules for chemical structures with potential of binding to a target receptor and prioritizing the most promising ones. Only a few of these will be selected for biological evaluation and further refinement through chemical synthesis. Despite the accumulated knowledge by pharmaceutical companies that continually improve the process of finding new drugs, a myriad of factors affect the activity of putative candidate molecules in vivo and the propensity for causing adverse and toxic effects is recognized as the major hurdle behind the current "target-rich, lead-poor" scenario. In this study we evaluate the use of several Machine Learning algorithms to find useful rules to the elucidation and prediction of toxicity using ID and 2D molecular descriptors. The results indicate that: i) Machine Learning algorithms can effectively use ID molecular descriptors to construct accurate and simple models; ii) extending the set of descriptors to include 2D descriptors improve the accuracy of the models.