Machine learning driven simulation of protein folding atomistic trajectories


The folding of proteins is an important biological process that determines the structure, role and functionality of proteins. It is often studied by molecular dynamics (MD) simulations, in order to obtain the folding trajectory of all the atoms in the system.
To date, pure MD simulations require huge computational resources and are still unable to access the timescales of folding processes that have biological relevance.
In my work, I am exploiting machine learning techniques and one recent AI milestone, Deepmind’s Alphafold, in order to create an advanced algorithm able to explore the folding trajectories within short computational times. It becomes possible to extract atomistic conformations from the folding pathways, and identify folding intermediates and long-lived states.
This method can be used to facilitate the identification of biologically relevant protein conformations, later to be used for pharmacological targeting or biophysical studies.



Sponsored by