Research And Implementation Of Drug Molecular Dynamics Simulation And Parallel Virtual Screen In Container Environment

With the strong computing power of High Performance Parallel Computing,the molecular dynamics simulation method of drug design based on the micro level can produce rich protein conformations,and the virtual screening method for drug design based on molecular docking can screen out the drug molecule with strong targeting,both of which can accelerate the development of new drugs.But for a long time,they have high requirements for the computing environment,and they rely on a large number of software,deployment is complex,cannot be migrated,and upgrade on the shared computing resources is prone to conflict with other software,which not only consumes a lot of time for pharmaceutical researchers,but also brings heavy burden to the deployment and maintenance of the computing cluster.In addition,there are a large number of compounds in the drug database,and the problem of long-term screening time also needs to be solved.This paper explores the use of container virtualization with isolation,security and migration as performance characteristics to solve the above problems.First,the performance loss differences between KVM,Docker and Singularity in CPU processing,storage,disk I/O and network bandwidth are studied by using standard benchmark programs such as HPL,HPCG,OSU and so on.Then,two typical HPC programs(VASP and WRF)are used in High Performance Computing to evaluate the performance of container in multi-node parallel processing,GPU computing and cross-platform migration under real-world examples.Next,the container with the best performance in the above benchmark performance experiment is selected to construct three kinds of container environments that can simultaneously satisfy three classical molecular dynamics simulation: NAMD(CPU and GPU Heterogeneous Computing),AMBER(GPU parallel version)and GROMACS(multi-node parallel version),which were applied to the molecular dynamics simulation of GLP-1R(type ? diabetes receptor),BRD9 protein(related to cancer),and anesthesia ?-opioid receptor with corresponding ligand.Finally,this paper studies the performance differences betweenthe compiled and precompiled versions of autodock Vina virtual screening tools,based on the optimal version of Vina,a multi-node parallel virtual drug screening container system based on MPI was constructed and deployed.Its parallel processing performance and the virtual drug screening of MTH1 kinase targeting tumor cells were studied.The experimental results show that the performance of Singularity container is better than KVM and Docker in computing peak,storage,network and disk I/O through benchmark performance evaluation,and it can achieve near-native computing results;and in the real application scenario,the performance of Singularity container fully meets the requirements of High Performance parallel Computing environment such as MPI,GPU and Infiniband,and can obtain the same accurate results as physical machine,the migration performance experiment also demonstrates the high performance stability of Singularity.In addition,the molecular dynamics simulation results of GLP-1R,BRD9 and ?-opioid receptors with corresponding ligands using container computing environment are closely to physical machines,and their performance loss is negligible;in the parallel virtual screening experiment of drugs targeting MTH1 kinase,the compiled version of Vina has the best performance,and the parallel virtual screening container system based on it has greatly reduced the drug screening time with the increase of the number of nodes,and the final screened drug molecule can be accurately docked on the active site of MTH1.All the images made in this paper have been uploaded to the image resource repository,which can be downloaded and used by drug researchers.