Mobile Menu

Rapid ML Screening Identifies Potential Drug Compounds For SARS-CoV-2

At the time of writing, there have been 137 million cases and nearly 3 million deaths from COVID-19. Although some targets have been identified for the treatment of COVID-19, there is still a lack of specific drugs that target the virus. In a recent study, researchers used rapid ML screening of existing drug compounds that could be used to treat SARS-CoV-2.

Support Vector Machine Models and Pseudotyped Particles

The study used a support vector machine (SVM) classification model with molecular fingerprinting to identify potential compounds that could aid drug discovery. An important factor in the infectiousness and pathogenesis of coronavirus is its entry into host cells. By blocking entry into the cell, the virus can no longer replicate and spread.

SARS-CoV-2 pseudotyped particles (PP) which are non-replicating versions of the virus, can be used in rapid ML screens and reporter assays. PPs are used as non-virulent versions of a virus, containing reporter RNA instead of the viral genome. These are safer and can be used in biosafety level 2 labs, rather than 3 for regular SARS-CoV-2.

Molecular Fingerprinting (FP)

Molecular fingerprints are being used widely in drug discovery as a way of encoding the structure of a molecule, which can be virtually screened for utility. There are many ways to encode these molecules, but the two main methods are based on 2D or 3D structural information. 3D structures are more computationally challenging to encode but provide important spatial context. The team used atom type molecular descriptors, which are 2D representations, and 3D atom-pair FPs for this study. Each method complements the weakness of the other and achieved a boost in predictivity by combining their strengths. It reached an area under the curve of the receiver operating characteristic (AUC-ROC) of 0.91.

Rapid ML Screening of NCATS libraries

The SVM model had been trained using the PP entry assay, cytotoxic assay and the two types of FP. This consensus model was then applied to three NCATS libraries, totalling 173,898 compounds. These had not been previously experimentally screened.

Of these, 255 were selected for experimental confirmation. From these 255, 116 were shown to have inhibitory activities on SARS-CoV-2 PP entry.


Although the exact method of how these compounds inhibit PP entry is yet to be determined, it would result in the inhibition of membrane fusion, preventing cell surface receptor binding.

The researchers concluded that the 116 compounds, serve as, “a great starting point for COVID-19 drug discovery.” And as only one FDA approved drug exists so far, there is plenty of room for more!

More on these topics

Covid-19 / drug discovery / Machine Learning

Share this article