Study On Conformational Search Optimization Algorithm Of Autodock Vina Suite And Cloud Platform Construction For Drug Screening

Lead compounds refer to compounds that are found to interact with a target in a disease during drug development.Once a lead compound is identified,drug development focuses on optimizing this compound to ultimately develop a drug with ideal efficacy and good safety.Virtual screening provides a new way to discover lead compounds,and molecular docking is a structure-based virtual screening method.However,the lengthy computational time required for molecular docking limits its utility for large-scale virtual screening and requires significant computing resources.To address this issue,an improved conformational search method was developed to reduce computational time.A cloud-based molecular docking platform was also developed to leverage significant cloud computing resources to enable parallel molecular docking.Approach in summary.(1)The improved quasi-Newton and line search methods were used for local search in Autodock Vina and its derivatives,while step length and descent direction pruning conditions were added to enable timely exit from local search and improve search speed.Increased simulated annealing iterations were also employed to improve search accuracy.The molecular docking programs were updated for one type of CPU and three types of GPU calculations and experimental tests were conducted.The results showed that the improved algorithm increased the speed by 272%,176%,190%,and 167% on the Autodock-GPU dataset,and 132%,136%,and 132% on the Drug Bank dataset,compared to the original algorithms.Furthermore,the average score and hit rate and enrichment factor were better than the original algorithm in most Ch EMBL datasets.(2)Dynamic allocation and fixed allocation algorithms were proposed for task allocation in the cloud-based molecular docking platform.These algorithms ensured efficient task execution while fully utilizing computing resources.9125 molecules in Drug Bank were docked on the RIPK1 target,and the dynamic allocation algorithm used the Quick-Vina2 program,800 CPUs,which was 97.15 times faster than a single node with 8 CPUs.The fixed allocation algorithm uses Vina-LBFGS-GPU+,5 graphics cards,which was 4.71 times faster than the node speed of a single graphics card.This paper describes algorithmic improvements and cloud platform designs for molecular docking,which improve the efficiency of drug screening using molecular docking and have practical applications for virtual screening.