Probing the protein space for extending the detection of weak homology folds

Redundancy of prediction methods has been used to explore the occurrence of weak homology protein motifs. A hybrid template-based algorithm has been implemented to predict different layers of protein structure by detecting domain building sub-structures, which share low sequence identity. Physicochemical determinants, secondary structure profiles, and multiple alignments have been analyzed to generate a broad set of structural sub-domains. Then, intensive computing procedures generated all the various tridimensional folds on the basis of secondary structure predictions, fragment assembly and detection of structural homologs. The proposed algorithm not only identifies common protein sub-structures, but also detects higher order architectures such as domain superfamilies/superfolds by linking backbone trajectories of supersecondary structures. Applying rigid transformation protocols, population of the detected domain building models with an average root mean square deviation from native structures of 2.3 Å and an average template modeling score from native structures of 0.43 has been obtained. The fold detection algorithm here proposed yields more accurate results than previously proposed methods, predicting structural homology also for proteins sharing less than 20% sequence identity. Our tools are freely available at http://www.acbrc.org/tools.html.

► We attempted to extend the detection of protein remote homology folds. ► Models detected share both high accuracy and low sequence identity. ► Our algorithm could be a complementary tool for exploring protein motifs. ► Motifs detected by our algorithm often share sequence identity lower than 20%. ► Domains with remote homology were detected by the merging of motifs trajectories.