Using Cyclica’s technology to identify repurposed drug candidates for COVID-19

At the time of writing this article, over 1.3 million people have been confirmed to be infected with the SARS-CoV-2 virus (with many, many more unconfirmed) and nearly 75,000 people have died from COVID-19 related illness. Countries the world over are enacting measures to slow the spread of the virus, hoping to relieve the burden on already taxed medical systems. Our health professionals are searching urgently for ways to help treat those affected, relying on the methods and medicines that they have at their disposal; unfortunately, given the novelty of the virus, no existing medicine is proven to treat and cure the afflicted. Vaccines and treatments specific to the virus will take too long to impact the current epidemic, though they are critical for long term measures. Only one drug-based strategy bears potential in the near term: drug repurposing. Given our strong background in molecular polypharmacology, at Cyclica, we applied our deep learning platform, MatchMaker, to the identification of FDA-approved drugs and experimental medicines being evaluated in the clinics that have potential to treat COVID-19. The findings of our study are now published in Chemrxiv. This work was the result of the tireless efforts of Cyclica’s R&D and Applied Science teams, and we hope it will contribute meaningfully to the search for COVID-19 therapies. 

To enable this effort, we assembled PolypharmDB, a new resource that contains over 10,224 approved and experimental medicines along with the computed list of all proteins predicted to interact with them. These predicted interactions were generated with Matchmaker, our AI-augmented, physics-based approach to predicting compatibility between a small molecule and a protein pocket on a proteome-wide level. These previously unanticipated interactions between a drug and a protein can motivate testing drug candidates for unconventional uses, such as COVID-19. The newly assembled PolypharmDB can be searched rapidly for medicines that are likely to interact with key proteins linked to SARS-CoV-2 infection, identifying candidate drugs as possible therapeutics. Critically, given the shifting understanding around SARS-CoV-2, PolypharmDB can be queried as new drug targets are discovered, returning additional drug candidates within a single business day. For example, in this study we honed in on two recently identified host targets, TMPRSS2 and cathepsin B, both of which are hypothesized to impact the life-cycle of SARS-CoV-2. 

In addition to TMPRSS2 and Cathepsin B, leveraging PolypharmDB we explored other host and viral protein drug targets, each with their own challenges. The host proteins selected for our initial investigation are linked to biology surrounding SARS and MERS infections; however, the role of these targets in SARS-CoV-2 is not well understood. Further, while MatchMaker demonstrates generalizability across proteins, it was trained on human drug-target interactions solely; we have not validated its performance on viral proteins. Despite these limitations, we believe that our approach is robust to some level of uncertainty and have included the results reported in our Chemrxiv manuscript for evaluation.

Figure 1: Disopyramide with predicted poses within the TMPRSS2 binding site. The area highlighted in yellow shows a 5 A radius surrounding the docked molecule. The reference ligand that was used to define the binding pocket from the crystal structure is displayed in gray.

Ultimately, PolypharmDB is a unique tool to evaluate off-target interactions of approved drugs and candidates, allowing us to uncover opportunities to repurpose drugs for novel indications rapidly. However, given that these are unorthodox applications for these medicines, the proposed drug candidates should be assessed for their relevance and re-tested in relevant patient populations.

Excitingly, adjacent to our work with PolypharmDB and repurposing known therapies as COVID-19 treatments, we applied our platform to novel drug discovery. Recently we took part in the Postera.ai challenge, which is open to all researchers and drug discovery scientists to design inhibitory molecules based on the XChem fragment study of 3CL-protease. Read more about this submission and our philosophy here.