3D-QSARpy: Combinando estratégias de seleção de atributos e técnicas de aprendizado de máquina para construir modelos QSAR 3D

Abstract

Quantitative Structure Activity Relationship (QSAR) is a technology in the field of medicinal chemistry that seeks to clarify the relationships between molecular structures and their biological activities. For this, QSAR models are constructed from the structural data (1D, 2D, 3D or 4D) from a series of molecules already tested for a given activity. Through predictions made by these models, it is aimed to identify which modifications in the molecule can influence, reinforcing or not the biological response. Such technology allows accelerating the development of new compounds by reducing the costs for drug design. Considering the briefly exposed context, the present work aims to develop a methodology for predicting biological activity in bioactive molecules. The methodology was successfully validated through the application of the tool in two sets of data, which results outperformed those previously published. The first set involving diabetes treatment, it reached r2 pred=0.91. The second set referring to cancer treatment, with r2 pred=0.98. Finally, two applications of the tool were performed, contributing to the identification of new bioactive molecular structures using different approaches. The first of which is intended for the treatment of chagas disease, including the construction of hybrid QSAR models for three series, obtaining r2 pred = 0.8, 0.68 e 0.85. The second application was the construction of QSAR-4D for the tuberculosis treatment with r2 pred = 0.72. It doesn’t matter if the experiments were for validation or for the identification of these new molecules. All of them demonstrated not only the efficiency of the proposed methodology and the developed tool, but also the versatility of possible applications with this methodology, either following its general pipeline or using it in a partially way combined with other existing tools.