Assessing the quality of folding and docking poses


The ability of discriminating the native pose of a protein among a large set of incorrect ones (decoys) is a fundamental step in any protein structure prediction process. This is usually accomplished via a scoring function, which associates to every pose a free energy value (score) whose minimum should always correspond, in principle, to the native pose. Since the physical laws underlying protein folding and protein-protein interaction phenomena are the same, we believe a single method for treating both problems could be devised.



BACH: Bayesian Analysis Conformation Hunt

BACH is a statistical knowledge-based potential trained on a set of ~450 folded globular monomers. It makes use of a pairwise, residue-based energy term, an atomwise clash term and a solvation contribution. The pairwise term has been recently refined including a difference between polar and apolar contacts (BACH-SixthSense). Unlike the best performing state-of-the-art scoring functions, BACH-SixthSense has proven able to reach top performances in both protein folding (CASP, fisa, 4state, ...) and protein-protein interaction (CAPRI, ZDOCK-generated decoy sets) tests. Here we report a comparison of the performance of BACH on 33 CASP and 14 CAPRI decoy sets. For each decoy set we sort the scoring functions energy values from the lowest to the highest and see at which point of the rank the native pose is found. The lower the rank, the better the performance.


Recently, BACH has also been adapted for working as a collective variable in enhanced sampling methods (cfr. metadynamics).


Bibliography