(589b) Loop Modeling in Proteins Using a Configurational Bias Approach

Loop modeling is important in homology modeling, enzyme design, and flexible backbone docking. Despite improvements in protein structure prediction in general, loop modeling remains very challenging, especially due to the difficulties of sampling diversely while maintaining the loop connectivity. Current approaches in loop modeling often only treat loops up to 8 or 12 residues, and the time required to model these loops is long. In this talk, we will present a new loop modeling algorithm based on a configurational bias approach, similar to that used in polymer modeling, where loops are deleted and then re-grown one residue at the time from both ends, alternating between the N- and C-terminus of the loop. The new algorithm reduces the time required to model loops by as much as five-fold. We found that this approach increases the sampling near the native conformation and therefore we are able to find the native conformation without requiring as much sampling (usually near-native structures can be found within the first 100-200 structures). Using the re-growing procedure, preliminary data indicate that for certain loops, the lowest energy structure is as much as 1.2 Å rmsd lower than previously published results for 12-residue loops and much less sampling is needed to achieve these high resolution results. On a small set of 12-residue loops from 10 proteins, the average rmsd of the lowest energy structure was 1.1 Å after only generating 100 structures for each protein. The efficiency and accuracy of the new technique promises to allow more aggressive sampling of flexible loop conformations in applications such as enzyme design and docking of antibodies.