Pandemic drugs at pandemic speed: infrastructure for accelerating COVID-19 drug discovery with hybrid machine learning- and physics-based simulations on high-performance computers

Abstract

The race to meet the challenges of the global pandemic has served as a reminder that the existing drug discovery process is expensive, inefficient and slow. There is a major bottleneck screening the vast number of potential small molecules to shortlist lead compounds for antiviral drug development. New opportunities to accelerate drug discovery lie at the interface between machine learning methods, in this case, developed for linear accelerators, and physics-based methods. The two in silico methods, each have their own advantages and limitations which, interestingly, complement each other. Here, we present an innovative infrastructural development that combines both approaches to accelerate drug discovery. The scale of the potential resulting workflow is such that it is dependent on supercomputing to achieve extremely high throughput. We have demonstrated the viability of this workflow for the study of inhibitors for four COVID-19 target proteins and our ability to perform the required large-scale calculations to identify lead antiviral compounds through repurposing on a variety of supercomputers.

Structures of the SARS-CoV-2 spike glycoprotein and applications for novel drug development

“…Top scoring 187 hits (75 FDA-approved) were further validated by all atom docking studies, and important molecular descriptors and promising chemical fragments are identified to guide future experiments. Coveney et al designed a novel in silico method for drug design by coupling ML with physics-based (PB) simulations ( Bhati et al, 2021 ). The accurate PB simulations would make the drug design process smarter by calculating the binding free energies of obtained hits from the output of a deep learning (DL) algorithm, which will then fed back to the DL algorithm to improve its predictive performance.…”

Which corners to cut? Guidelines on choosing optimal settings to maximise sampling with limited computational resources

“…Clear evidence of non-Gaussian distributions typically requires substantial data sets 9 and thus ensembles of much larger size than the ones we recommend for routine (non-corner-cutting) uncertainty quantification, namely with 25 replicas in ESMACS and 5 for TIES. We have used both relaxed and full ensemble protocols in our IMPECCABLE workflow as filters to perform screening at different stages with different levels of precision 10 . Such relaxed protocols are termed as "coarse-grained (CG)" against the full protocol that is termed as "fine-grained (FG)".…”

Alchemical Free Energy Estimators and Molecular Dynamics Engines: Accuracy, Precision, and Reproducibility

“… 11 Moreover, these calculations can allow much larger areas of chemical space to be explored than would be possible experimentally. Compounds drawn from this chemical space can be selected from numerous sources such as chemical libraries, 12 repurposing of approved drugs, 13 generative AI methods, 14 − 16 or even other free energy calculations. 17 …”