GPU Implementation of the Flexible Block Method for Normal Mode Analysis
Normal Mode Analysis (NMA) decomposes a protein's movement by frequency. NMA can be used to identify important dominant movements, identify mechanisms of allosteric effects, and identify effects of binding with ligands. However, it requires
diagonalizing a large 3N x 3N Hessian matrix. Full diagonalization of such large matrices is very computationally expensive, so the Flexible Block Method (FBM), which extends the Rotation-Translation Block Method (RTB) developed by Tama, et al, was developed. FBM is able to approximate the slowest normal modes of a protein accurately and quickly. In this video, OpenMM 2012 Visiting Scholar RJ Nowling describes his project to implement FBM entirely on GPUs for use with OpenMM and his experiences as a visiting scholar.