Parts | Days | Selection | Search | Downloads | Help
BP: Fachverband Biologische Physik
BP 7: Poster I
BP 7.28: Poster
Monday, March 23, 2009, 17:45–20:00, P3
Improving protein structure prediction using sequence-derived structure profiles — •Katrin Wolff1, Andrea Cavalli2, Michele Vendruscolo2, and Markus Porto1 — 1Institut für Festkörperphysik, TU Darmstadt, Germany — 2Department of Chemistry, University of Cambridge, Cambridge, UK
A crucial step in the prediction of protein structures is the transition from low- to high-resolution models. There exist various tools that generate candidate sets that contain high-quality, yet coarse-grained, structures. In a subsequent refinement step these structures are improved to all-atom representations and minimized using a high-resolution energy functional. Due to limited computer time it is vital to restrict this refinement step to promising candidates and to identify the best structures. The energy functional used in the structure generation step, however, is only of limited use for the problem of selecting these ‘good’ structures. We discuss the use of structure profiles for this filtering step. As a proof of principle we show that the exact profile (derived from the native structure) is very reliable in choosing candidates with low RMSD to the native structure and clearly outperforms other filtering methods like filtering by energy or clustering the decoy set. Such structure profiles can be predicted to good accuracy from sequence [1,2]. We therefore explore the use of profiles as predicted from sequence and show that for sufficiently high accuracy this approach is also superior to the other methods of filtering.
[1] A. R. Kinjo et al., BMC Bioinformatics 7, 401 (2006).
[2] J. Minning, F. Teichert, U. Bastolla, M. Porto, in preparation.