Novel methods for the acceleration of protein simulation.

Abstract

Molecular dynamics (MD) simulations of many biochemically relevant timescales are still beyond the limits of state-of-the-art supercomputers. Analog electronic circuits offer an alternative method of computation to the classical computers used today. These computers are uniquely performant at solving systems of equations such as those that describe physical systems, making them ideally suited to MD simulations. Thanks to recent advancements in semiconductor fabrication, analog computing has seen a resurgence, particularly in computationally intensive sectors such as artificial intelligence. The aim of this work is to explore the application of analog computing to the acceleration of MD simulations. We present a ground-up approach for building custom electronics circuits that implement the force fields classically used to simulate biomolecular systems. Our design incorporates digital components to mitigate against undesirable effects such as noise and drift, which arise from the environmental factors to which analog circuits are sensitive. This hybrid approach also enables us to impart the system with the ability to self-calibrate and self-adjust. We have designed and simulated circuits that perform MD simulations of a small peptide at timescales that exceed those that are classically feasible. In addition, we have produced physical proof-of-concept circuit boards that substantiate the results of those simulations. Our findings suggest analog-based simulators may provide a viable route to long-timescale simulation with a significant reduction in associated power consumption.