COVID-19 in-silico research to identify
tentative anti-viral treatments

Case study

Machine learning in bioinformatics provides new means to conduct advanced biomedical research. One emerging application of the technology is in expediting drug discovery. In-silico studies can use as data the molecular structure of biomolecules that are drug targets (e.g., enzymes in key metabolic pathways) and their associations with pharmaceuticals. Neural networks trained with these datasets can then predict how diverse molecular structures can associate with each other, which streamlines the identification of compounds that may have therapeutic potential and thus merit further study using in-vivo and in-vitro laboratory techniques.

At the onset of the pandemic, we at Zetane were able to use public datasets of antiviral drugs and their association with viral proteins to train two convolutional neural networks. We produced two models that suggested tentative anti-viral treatments for COVID-19; some of the drugs identified by our models were identified as tentative therapeutics by independent clinical trials and in-vitro investigations. Here we present a multimedia article, research article and video clip from our project. Note that a particular advantage of the Zetane Engine is the ability to see how the intermediate layers of our neural networks are indeed identifying specific amino acid sequences of viral proteins and their associations with antivirals. This demystifies the training process, where it is not some random process; rather, the training is precise and ordered. Such observations facilitate necessary peer-review and regulatory tasks for our study.

Our in-silico research findings are now in the process of further investigation using in-vitro investigations. Sign up for our newsletter in the top menu to know when our results are ready for public scrutiny.