Drugsniffer: An Open Source Workflow for Virtually Screening Billions of Molecules for Binding Affinity to Protein Targets

Abstract

The SARS-CoV2 pandemic has highlighted the importance of efficient and effective methods for identification of therapeutic drugs, and in particular has laid bare the need for methods that allow exploration of the full diversity of synthesizable small molecules. While classical high-throughput screening methods may consider up to millions of molecules, virtual screening methods hold the promise of enabling appraisal of billions of candidate molecules, thus expanding the search space while concurrently reducing costs and speeding discovery. Here, we describe a new screening pipeline, called drugsniffer, that is capable of rapidly exploring drug candidates from a library of billions of molecules, and is designed to support distributed computation on cluster and cloud resources. As an example of performance, our pipeline required ∼40,000 total compute hours to screen for potential drugs targeting three SARS-CoV2 proteins among a library of ∼3.7 billion candidate molecules.

Advances and critical assessment of machine learning techniques for prediction of docking scores

“…Precision‐recall curves and receiver operating characteristic curves (ROC) are shown in Figure S13, choosing threshold value of −8.6 kcal/mol to define the required properties. In the work of Venkatraman et al [28], with 1 000 000 docked compounds, the receiver operating characteristic curve (ROC) is performing worse compared to ours for the in‐vitro‐only set. The authors further report AUC value of 0.746 [28], while the TF, XG, and SP models lead to AUC values of in‐vivo and in‐vitro‐only sets of 0.964, 0.972, 0.977 and 0.943, 0.939, 0.955, respectively, see Figure S13.…”

“…In the work of Venkatraman et al [28], with 1 000 000 docked compounds, the receiver operating characteristic curve (ROC) is performing worse compared to ours for the in‐vitro‐only set. The authors further report AUC value of 0.746 [28], while the TF, XG, and SP models lead to AUC values of in‐vivo and in‐vitro‐only sets of 0.964, 0.972, 0.977 and 0.943, 0.939, 0.955, respectively, see Figure S13. However, it should be noted that ROC curve depends on the choice of the threshold, that is, in Venkatraman et al [28] the authors work with an active: decoy ratio of 3900:213000 (3900 ligand‐protein pairs and 213 000 decoy‐protein pairs).…”

Large-Scale Docking in the Cloud

“…Regarding cloud-based docking, there have been several studies that were encouraging. For instance, some groups have published docking pipelines suitable for running in the cloud. ,,− Although available as open-source, these generally lack a step-by-step protocol suitable for nonexperts. Commercial solutions to docking in the cloud are also available, such as Orion (OpenEye, Santa Fe NM).…”

Large Scale Docking in the Cloud

“…For instance, some groups have published docking pipelines suitable for running in the cloud 4,7,[14][15][16] . Although available as open source, these generally lack a step-by-step protocol suitable for non-experts.…”