Scientists from TCS Innovation Labs in Hyderabad are harnessing the power of artificial intelligence (AI) to identify new molecules which might have the potential to target specific parts of the novel coronavirus (SARS-CoV-2). Using new methods, they have identified 31 candidate small molecules, which may serve as inhibitors of the chymotrypsin-like protease, one of the key drug targets in the fight against COVID-19.
The results have been posted in a preprint repository ChemRxiv. Preprints are yet to be peer-reviewed and published in scientific journals.
The genome of the novel coronavirus codes for several proteins that have crucial roles in entry of the virus into the host cell, its replication, assembly and host-pathogen interactions. Some of these proteins that help the virus perform its functions are common targets for drug developers. Among these drug targets are the spike protein, which helps the virus attach itself to the host cell and enter it, and viral proteases which help it replicate.
Role of viral protease
“The viral RNA synthesises two long polyproteins when it infects human cells via a human cell surface protein. The role of the protease protein is to cut the polyproteins to individual proteins, so that new viruses can be assembled. This is important for its replication and survival,” explains Arijit Roy from the Life Sciences Division of TCS Innovation Labs, Hyderabad, in an email to The Hindu.
The chymotrypsin-like protease or the main protease primarily does the function of cleaving the polyprotein into proteins and the papain-like protease also aids in this process. The former was chosen as the drug target by the group for their research.
First, using a database of approximately 1.6 million drug-like small molecules from the ChEMBL database, the researchers trained the generative deep neural network model. As a second step, the network was re-trained with protease inhibitor molecules. This was done with a view to narrow the focus of the neural network on to a smaller subset of the chemical space. “We trained the system with all available protease inhibitors and asked the pre-trained model to produce more new molecules that possess the characteristics of protease inhibitors,” says Dr. Roy, who is an author of the preprint. “Finally, we checked how well these newly produced molecules can bind to the target protein — chymotrypsin-like protease of the virus.”
Starting from a space of nearly 50,000 molecules, the team has made a short list of 31 candidates. Two of the designed molecules had a high degree of similarity to Aurantiamide, a naturally occurring antiviral-compound.
“Our aim was to create new molecules which possess the characteristics of protease inhibitors. We checked, whether these molecules retain all the drug-like properties. We also checked how easily they can be synthesised. All these were part of the AI-based model,” says Dr. Roy.
Drug discovery is a complex process, needing several layers of validation before the drug may come in use. In this work, the researchers have brought down the time taken for the initial step of designing suitable candidate molecules for testing from years to just a week, reinforcing the power of AI in handling huge datasets.
“TCS has signed an MoU for collaboration with CSIR. The clinical trials will take time. The first step is the chemical synthesis and biological testing in vitro, followed by pre-clinical testing on laboratory animals,” he says.