Given the scale and rapid spread of the 2019 novel coronavirus (2019-nCoV) acute respiratory disease, there is an immediate need for medicines that can help before a vaccine can be produced. Results of rapid sequencing of 2019-nCoV, coupled with molecular modelling based on the genomes of related virus proteins,1 have suggested a few compounds that are likely to be effective, including the anti-HIV lopinavir plus ritonavir combination.
BenevolentAI's knowledge graph is a large repository of structured medical information, including numerous connections extracted from scientific literature by machine learning.2 Together with customisations bespoke to 2019-nCoV, we used BenevolentAI to search for approved drugs that could help, focusing on those that might block the viral infection process. We identified baricitinib, which is predicted to reduce the ability of the virus to infect lung cells.
Most viruses enter cells through receptor-mediated endocytosis. The receptor that 2019-nCoV uses to infect lung cells might be ACE2, a cell-surface protein on cells in the kidney, blood vessels, heart, and, importantly, lung AT2 alveolar epithelial cells (figure ). These AT2 cells are particularly prone to viral infection.3 One of the known regulators of endocytosis is the AP2-associated protein kinase 1 (AAK1). Disruption of AAK1 might, in turn, interrupt the passage of the virus into cells and also the intracellular assembly of virus particles.4
Figure.
The BenevolentAI knowledge graph
The BenevolentAI knowledge graph integrates biomedical data from structured and unstructured sources. It is queried by a fleet of algorithms to identify new relationships to suggest new ways of tackling disease. 2019-nCoV=2019 novel coronavirus. AAK1=AP-2 associated kinase 1. GAK=cyclin g-associated kinase. JAK1/2=janus kinase 1/2.
Of 378 AAK1 inhibitors in the knowledge graph, 47 have been approved for medical use and six inhibited AAK1 with high affinity. These included a number of oncology drugs such as sunitinib and erlotinib, both of which have been shown to inhibit viral infection of cells through the inhibition of AAK1.5 However, these compounds bring serious side-effects, and our data infer high doses to inhibit AAK1 effectively. We do not consider these drugs would be a safe therapy for a population of sick and infected people.
By contrast, one of the six high-affinity AAK1-binding drugs was the janus kinase inhibitor baricitinib, which also binds the cyclin G-associated kinase, another regulator of endocytosis. Because the plasma concentration of baricitinib on therapeutic dosing (either as 2 mg or 4 mg once daily) is sufficient to inhibit AAK1, we suggest it could be trialled, using an appropriate patient population with 2019-nCoV acute respiratory disease, to reduce both the viral entry and the inflammation in patients, using endpoints such as the MuLBSTA score, an early warning model for predicting mortality in viral pneumonia.7
This online publication has been corrected. The corrected version first appeared at thelancet.com on June 18, 2020
Acknowledgments
JS is editor-in-chief of Oncogene. JS has previously sat on a number of scientific advisory boards, including BenevolentAI, and has consulted with Lansdowne partners, Vitruvian, and Social Impact Capital; he now sits on the Board of Directors for BB Biotech Healthcare Trust and chairs Xerion Healthcare. All other authors are employees of BenevolentAI. Events in relation to the 2019-nCoV outbreak are evolving rapidly, and we make our initial thoughts available in this Correspondence in good faith and to assist in the global response. Our early investigations and suggestions require further detailed work and analysis and should not be relied on as constituting any kind of medical or other advice or recommendation.
References
- 1.Gruber CC, Steinkellner G. Wuhan coronavirus 2019-nCoV—what we can find out on a structural bioinformatics level. Jan 23, 2020. https://innophore.com/2019-ncov/
- 2.Segler MHS, Preuss M, Waller P. Planning chemical syntheses with deep neural networks and symbolic AI. Nature. 2018;555:604–610. doi: 10.1038/nature25978. [DOI] [PubMed] [Google Scholar]
- 3.Zhao Y, Zhao Z, Wang Y. Single-cell RNA expression profiling of ACE2, the putative receptor of Wuhan 2019-nCov. BioRxiv. 2020 DOI:https://doi.19985 published online Jan 26. [Google Scholar]
- 4.Lu R, Zhao X, Li J. Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding. Lancet. 2020 doi: 10.1016/S0140-6736(20)30251-8. published online Jan 30. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Pu SY, Xiao F, Schor S. Feasibility and biological rationale of repurposing sunitinib and erlotinib for dengue treatment. Antiviral Res. 2018;155:67–75. doi: 10.1016/j.antiviral.2018.05.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Chen N, Zhou M, Dong X. Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in Wuhan, China: a descriptive study. Lancet. 2020 doi: 10.1016/S0140-6736(20)30211-7. published online Jan 29. [DOI] [PMC free article] [PubMed] [Google Scholar]
Uncited Reference
- 6.Sorrell F, Szklarz M, Abdul Areez KR. Family-wide structural analysis of human numb-associated protein kinases. Structure. 2016;24:401–411. doi: 10.1016/j.str.2015.12.015. [DOI] [PMC free article] [PubMed] [Google Scholar]