Based on AutoSite, AutoSite2 is developed to better predict pockets for both peptide ligands and small molecule ligands. AutoSite2 exploits an estimated volume of the peptide ligand to predict binding sites that better overlap with the bound ligand. The core algo-rithm is to repeat the AutoSite algorithm while relaxing the cutoff values used for thresholding the affinity maps. As the thresholds become less stringent, the clusters of 3D points grow larger and start merging. AutoSite2 improves the peptide pocket quality described by Jaccard coefficient by 50% compared with AutoSite, AutoLigand and FPocket. AutoSite2 also achieve similar or better prediction accuracy with PeptiMap and PEP-SiteFinder while only using a fraction of computation time.
AutoSite2 is a part of MGLtools2 and a pre-release trial package is available.