Binding affinity estimation between protein and ligand is an essential part of the drug discovery process. Different approaches have been developed to build functions (empirical, forcefield based, knowledge based and, more recently, machine learning based) to evaluate the binding affinity between the receptor and the possible poses of the ligand generated by the docking algorithms. 

These functions are called scoring functions and they classify different protein-ligand complexes by returning a score related to free energy change value in consequence of docking. However, the number of ligands to be evaluated in virtual screening campaigns is usually so high that in order to obtain results in acceptable times, approximations are introduced that involve poorly accurate binding affinity predictions. 

Machine learning methods can be employed to address this problem. Recent studies have explored the usefulness of image recognition-based deep learning approaches to select the most promising ligands, as well as their most probable poses within a protein pocket. However, to generate a robust and reliable model for each complex, it must be trained with hundreds of poses of the same ligand. 

In this project, Exscalate aims to build a trained, validated and tested machine learning model, thanks to the computing power of the CINECA HPC infrastructure, which is capable of discriminating only the admissible poses of the ligand at hand for subsequent binding affinity evaluation. This will significantly reduce the execution time required for virtual screening campaigns.