DeeplearningVS is a project which aims to study a novel rescoring method based on deep learning (DL) techniques to enhance the accuracy of docking results, and boost the structure-based virtual screening outcome.

For machine learning in general, and even more for DL, an integral asset to success is the availability of a large training data set, as well as a meaningful feature encoding for the problem in hand. In this work, the increasingly large number of available protein-ligand complex data will be used and further enriched by a novel idea of artificially created analogues. Furthermore, different encoding strategies (grid or interaction based methods) will be further explored and combined with novel features to improve model performance.

Besides, thoroughly retrospective evaluation of the model quality, the complete docking and rescoring workflow will be implemented to screening for novel inhibitors of a target of interest together with our collaboration partners.