Homology Modeling With Internal Coordinate Mechanics: Deformation Zone Mapping and Improvements of Models via Conformational Search. PROTEINS: Structure, Function and Genetics

ABSTRACT Five models by homology con-taining insertions and deletions and ranging from 33 % to 48 % sequence identity to the known homologue, and one high sequence identity (85%) model were built for the CASP2 meeting. For all five low identity targets: (i) our starting models were improved by the Internal Coordi-nate Mechanics (ICM) energy optimization, (ii) the refined models were consistently better than those built with the automatic SWISS-MODEL program, and (iii) the refined models differed by less than 2 % from the best model submitted, as judged by the residue contact area difference (CAD) measure [Abagyan, R.A., Totrov, M.J. Mol. Biol. 268:678–685, 1997]. The CAD measure is proposed for ranking models built by homology instead of global root-mean-square deviation, which is frequently domi-nated by insignificant yet large contributions from incorrectly predicted fragments or side chains. We demonstrate that the precise identi-fication of regions of local backbone deviation is an independent and crucial step in the homol-ogy modeling procedure after alignment, since aligned fragments can strongly deviate from the template at various distances from the alignment gap or even in the ungapped parts of the alignment. We show that a local alignment score can be used as an indicator of such local deviation. While four short loops of the meet-ing targets were predicted by database search, the best loop 1 from target T0028, for which the correct database fragment was not found, was predicted by Internal Coordinate Mechanics global energy optimization at 1.2 Å accuracy. A classification scheme for errors in homology modeling is proposed. Proteins, Suppl. 1:29–37