During the last two decades, the structure-based virtual screening (SBVS) has become an important tool for rational design of drugs and agrochemicals. However, despite of many successful cases the SBVS still has some limitations, especially regarding the ranking of potential ligands according to their binding affinity. By analysing the protein-ligand interaction nano-environment space, we aim to identify the characteristics that are important and essential to the molecular recognition process and then create models that may improve the ranking in SBVS campaigns. By carefully selecting corresponding physical-chemical and structural features provided by STING_DB and STING_RDB databases, the protein-ligand interaction nano-environment space will be analysed using nonparametric machine learning methods. Given the success of similar approaches designed for other biological problems (protein-protein interface prediction, identification of catalytic site and binding site residues) by the Embrapa's Computational Biology Research Group, wherein this project is to be undertaken, and also by considering our preliminary results, we hope that it will be possible to derive models that can reliably and efficiently predict the affinity of small chemical compounds for protein targets.
News published in Agência FAPESP Newsletter about the scholarship: