A Local Rigid Body Framework for Global Optimization of Biomolecules

We present a local rigid body framework for simulations of biomolecules. In this framework, arbritrary sets of atoms may be treated as rigid bodies. Such groupings reduce the number of degrees of freedom, which can result in a significant reduction of computational time. As benchmarks, we consider global optimization for the tryptophan zipper (trpzip 1, 1LE0; using the CHARMM force field) and chignolin (1UAO; using the AMBER force field). We use a basin-hopping algorithm to find the global minima and compute the mean first encounter time from random starting configurations with and without the local rigid body framework. Minimal groupings are used, where only peptide bonds, termini, and side chain rings are considered rigid. Finding the global minimum is 4.2 and 2.5 times faster, respectively, for trpzip 1 and chignolin, within the local rigid body framework. We further compare O(10⁵) low-lying local minima to the fully relaxed unconstrained representation for trpzip 1 at different levels of rigidification. The resulting Pearson correlation coefficients, and thus the apparent intrinsic rigidity of the various groups, appear in the following order: side chain rings > termini > trigonal planar centers ≥ peptide bonds ≫ side chains. This approach is likely to be even more beneficial for structure prediction in larger biomolecules.