Word embedding mining for SARS-CoV-2 and COVID-19 drug repurposing

Background: The rapid spread of illness and death caused by the severe respiratory syndrome coronavirus 2 (SARS-CoV-2) and its associated coronavirus disease 2019 (COVID-19) demands a rapid response in treatment development. Limitations of de novo drug development, however, suggest that drug repurposing is best suited to meet this demand. Methods: Due to the difficulty of accessing electronic health record data in general and in the midst of a global pandemic, and due to the similarity between SARS-CoV-2 and SARS-CoV, we propose mining the extensive biomedical literature for treatments to SARS that may also then be appropriate for COVID-19. In particular, we propose a method of mining a large biomedical word embedding for FDA approved drugs based on drug-disease treatment analogies. Results: We first validate that our method correctly identifies ground truth treatments for well-known diseases. We then use our method to find several approved drugs that have been suggested or are currently in clinical trials for COVID-19 in our top hits and present the rest as promising leads for further experimental investigation. Conclusions: We find our approach promising and present it, along with suggestions for future work, to the computational drug repurposing community at large as another tool to help fight the pandemic. Code and data for our methods can be found at https://github.com/finnkuusisto/covid19_word_embedding.