Development and application of bioinformatics tools for protein loop modelling

Abstract

Prediction of protein loop structures is crucial for protein structure modelling, structural refinement, antibody design, or ion channels modelling. Here we propose several methodological improvements to our loop modelling online service: Random Coordinate Descent (RCD+ http://rcd.chaconlab.org/). This server combines an ab initio loop closure algorithm with a full-atom refinement in Rosetta. We propose to include a novel knowledge-based pairwise potential (KORP) which takes into account information of the relative position and orientation per residue. The inclusion of KORP, a Ramachandran potential correction, and an extensive parameter optimization significantly improve the prediction accuracy of the loop modelling procedure. Moreover, superior efficiency has been achieved by drastically reducing the number of loop candidates to be further refined. The approach has been successfully validated with several standard loop benchmarks. Interestingly, promising results were obtained even in one of the most challenging applications scenarios: the H3 loops prediction of the antibody complementary determining regions.