Large-Scale Docking in the Cloud

Abstract

Molecular docking is a pragmatic approach to exploit protein structures for new ligand discovery, but the growing size of available chemical space is increasingly challenging to screen on in-house computer clusters. We have therefore developed AWS-DOCK, a protocol for running UCSF DOCK in the AWS cloud. Our approach leverages the low cost and scalability of cloud resources combined with a low-molecule-cost docking engine to screen billions of molecules efficiently. We benchmarked our system by screening 50 million HAC 22 molecules against the DRD4 receptor with an average CPU time of around 1 s per molecule. We saw up to 3-fold variations in cost between AWS availability zones. Docking 4.5 billion lead-like molecules, a 7 week calculation on our 1000-core lab cluster, runs in about a week depending on accessible CPUs, in AWS for around $25,000, less than the cost of two new nodes. The cloud docking protocol is described in easy-to-follow steps and may be sufficiently general to be used for other docking programs. All the tools to enable AWS-DOCK are available free to everyone, while DOCK 3.8 is free for academic research.

DockOpt: A Tool for Automatic Optimization of Docking Models

“…DockOpt jobs using the default configuration can be expected to take up to a day for larger retrospective data sets of several thousands of molecules (Figure 10), though the exact time will depend on the server load as well as to what degree the parameter search algorithm converges to similar parameterizations over time. The resultant model can be deployed in a large-scale docking screen on any system with the necessary compute resources, such as a departmental cluster or via a cloud platform (e.g., AWS 54 ).…”

“…Second , the automated procedure can optimize the docking parameters significantly better than our previous automated system and, for most targets, produces docking models that are suitable for large-scale prospective screens. Third , the new pipeline can be installed locally or accessed via a new web interface we have created, and the resultant docking model may be downloaded and deployed for large-scale docking, either on premises or in the cloud . We take up each of these themes in turn.…”

Compromise in Docking Power of Liganded Crystal Structures of Mpro SARS-CoV-2 Surpasses 90% Success Rate

“…Nevertheless, the machine learning approach of Gentile et al (that was based on an iteratively updated molecular docking campaign on a million of compounds) was able to identify 117 compounds with inhibition activity against M pro of SARS-CoV-2 from a set of 40 billion compounds. In addition, billion compounds’ docking campaigns are reported to be realistically retrieved in a couple of days. , Nevertheless, an effective computational shrinking of the potential compounds upon molecular docking is to be considered with great care in the future when dealing with, e.g., inaccuracy of scoring functions and/or even wrong prediction of binding poses by the programs …”

HIt Discovery using docking ENriched by GEnerative Modeling (HIDDEN GEM): A novel computational workflow for accelerated virtual screening of ultra‐large chemical libraries

“…Indeed, it has been shown that VS of ultra‐large chemical libraries using molecular docking can identify greater number of diverse ligands with high binding affinity [5–10]. However, molecular docking is inherently computationally demanding, even when accelerated with recent advancements in hardware [11] and cloud computing [12,13]. Expanding upon estimates reported recently by Sadybekov et al.…”

“…affinity [5][6][7][8][9][10]. However, molecular docking is inherently computationally demanding, even when accelerated with recent advancements in hardware [11] and cloud computing [12,13]. Expanding upon estimates reported recently by Sadybekov et al [14], docking of the Enamine REAL Space would cost approximately $3,000,000 dollars using Amazon Web Service (AWS; https://aws.amazon.com/).…”