AlphaFold 2.x is a useful tool for anyone
working in structural biology or related fields. AlphaFold is great at
predicting structures of proteins together with confidence scores. Maybe less
well-known feature is that AlphaFold can be used also to predict probability
distributions of distances of all pairs of amino acid residues in the predicted
structure. We use this Alpha Fold output to infer the intramolecular potentials
acting on the protein of interest. Suchinformation combined with coarse-graining of the
molecule can be used to generate pseudo-trajectories of the protein, predicting
the dynamics, of the secondary structures, multi-domain interaction or
evaluating flexibility of intrinsically disordered parts of the structure. We
show how this approach works on several systems.
This work was supported by Czech Science Foundation (22-29667S) and by COST Action ML4NGP, CA21160, . Computational resources were provided by the e-INFRA CZ project (ID:90254), supported by the Ministry of Education, Youth and Sports of the Czech Republic.