Exploration of Low-Cost Long-Timescale Free Energy Perturbations on FPGAs

Drug discovery is challenging and expensiveespecially in the phase of lead optimization whichpredicts binding to a targetComputational approaches can add substantial valueComputingrelative binding free energiesRBFEbetween congeneric molecules using molecular dynamicsMDgreatly reduces the search space and considerably improves convergenceThe Silicon TherapeuticsSTXINSITE platform uses an RBFE protocol based on MD running on GPUs as part of the leadoptimization pipelineSTX has proven the capabilities of INSITE by finding small moleculeinhibitors for challenging targets that have no known small molecule inhibitorsThe two criticalbottlenecks are throughput and timescaleFor throughputwhile GPUs have substantially improvedcost effectivenessimprovement is still highly desirableespecially for energy cost persimulationFor timescalecurrent GPU RBFE simulations typically runns or less which oftenyields unconverged resultsWe propose to address both problems by accelerating INSITE withFPGA based clusters and FPGA cloud instancesFPGAs are commodity computation devices whose primaryuse has been in communication routersthey are also ideal for MDThe hardware adapts to theapplicationrather than the reverseand so effects high efficiencyAlsosince FPGAs are hybridcommunication computation processorslarge scale communication can proceed with both highbandwidth and low latencyOur preliminary work has shown that MD on clusters of FPGAs approachesthe performance of proprietary ASIC based clusters for several core functionsand that a currentnode FPGA accelerated cluster can simulate aK particle model at a rate of overus per daythis isx that of a commodity cluster of any sizeOur overall goal is to a create acommercial quality pipeline for running economical and long timescale RBFE simulationsTheprograms will be highly useful to the internal drug discovery efforts of STXSTX will also exploreavenues to provide the software as a service either in the cloude gthrough AWSor through anin house platformThe overall Aim of this Phase I proposal is to prove the concept ofFPGA accelerated RBFE for clouds and clusters in delivery of the modeled performancein validationof simulation qualityand in other software engineering metricsTherefore our Aimis acceleratekey RBFE functions with FPGAswith subtasks designed to span the space of target systems anddeployment scenariosOur Aimis to enhance the STX computational toolflow to create internaland external infrastructure to support the work in Aimto develop new computational methods totake advantage of the target architecturesand to evaluate and specify deployment options Computational approaches have been found to substantially aid in the discovery of new drugsThegoal of this proposal is to apply novel and emerging computational methods to improve thecost effectiveness and quality of these approaches