2019/09/17 by Hyungro Lee, Heng Ma, Lee, Hyungro +9 · 2 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · Materials Science · #Distributed #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning in Materials Science #Parallel #Protein Structure and Dynamics #Software Engineering Research #and Cluster Computing (cs.DC)
paper · pdf · doi:10.48550/arxiv.1909.07817
openalex publication_date 2019/09/17 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
Simulations of biological macromolecules play an important role in\nunderstanding the physical basis of a number of complex processes such as\nprotein folding. Even with increasing computational power and evolution of\nspecialized architectures, the ability to simulate protein folding at atomistic\nscales still remains challenging. This stems from the dual aspects of high\ndimensionality of protein conformational landscapes, and the inability of\natomistic molecular dynamics (MD) simulations to sufficiently sample these\nlandscapes to observe folding events. Machine learning/deep learning (ML/DL)\ntechniques, when combined with atomistic MD simulations offer the opportunity\nto potentially overcome these limitations by: (1) effectively reducing the\ndimensionality of MD simulations to automatically build latent representations\nthat correspond to biophysically relevant reaction coordinates (RCs), and (2)\ndriving MD simulations to automatically sample potentially novel conformational\nstates based on these RCs. We examine how coupling DL approaches with MD\nsimulations can fold small proteins effectively on supercomputers. In\nparticular, we study the computational costs and effectiveness of scaling\nDL-coupled MD workflows by folding two prototypical systems, viz., Fs-peptide\nand the fast-folding variant of the villin head piece protein. We demonstrate\nthat a DL driven MD workflow is able to effectively learn latent\nrepresentations and drive adaptive simulations. Compared to traditional\nMD-based approaches, our approach achieves an effective performance gain in\nsampling the folded states by at least 2.3x. Our study provides a quantitative\nbasis to understand how DL driven MD simulations, can lead to effective\nperformance gains and reduced times to solution on supercomputing resources.\n