Abstract

In the field of clinical research, it is crucial to determine the safety and efficacy of current drugs and further accelerate the discovery of new active compounds. The traditional methods performed expensive laboratory tests. These methods are very costly and require a long time. However, bioinformatics can greatly facilitate the clinical research for the above purposes, providing the prediction of drug toxicity and activity in new diseases and evaluating the validity of active compounds discovered in clinical trials. This can be achieved through the availability of bioinformatics tools and methods of computer virtual screening (VS) that allow the test all the necessary hypotheses before doing the clinical trials, such as structural docking using the BINDSURF program. However, the accuracy of most VS methods is severely restricted due to constraints in the affinity or scoring functions describing biomolecular interactions, and even today these uncertainties are not fully known. In this paper we address this problem by proposing a new approach in which neural networks are trained with information on databases of known compounds (target proteins and drugs), for further exploiting the method to improve the accuracy of VS BINDSURF predictions of affinity.