Project iGuess

A project of WP5, called iGUESS (Intelligent GUess for Electronic Structure Simulations), will construct a set of optimized basis functions for density functional theory (DFT) simulations and of an optimal initial density to be used in a self-consistency cycle. The team will investigate the possibility to use methods of data analytics and machine learning to achieve a significant performance boost of extreme scale ab-initio simulations in materials  science. These new machine learning methods open new possible research areas around life science, and more specifically about the covid 19 research. The applications developped within iGUESS, namely BigDFT, can be associated to highly accurate approaches of molecular dynamics (the Polaris polarizable force field code) in order to produce

populations of data that are able to address for dynamical quantities (solvent and temperature effects) as well as entropic contributions.

A concrete application will be the selection of inhibitors of the SARS-CoV-2 Main Protease by searching a panel of molecules for candidates targeting a given protein on the surface of a virus.

The objective is to go on a method as precise as possible in order to be able to do screening but also to serve as a reference for lighter docking methods which need very precise parameters to validate or complete their models. The generated inputs/outputs are potentially very large (a few GB for each of the snapshot analyzed – therefore potentially many Terabytes of data for the whole project

 – and this requires a suitable dataflow between the two codes, Polaris and BigDFT, as well as for the processing of fragments. Machine learning algorithms can be used to identify and target better which results should be exploited, leading to a large acceleration regarding the time to solution.