Accurate Identification of SARS-CoV-2 from Viral Genome Sequences using Deep Learning

Alejandro Lopez-Rincon; Alberto Tonda; Lucero Mendoza-Maldonado; Eric Claassen; Johan Garssen; Aletta D. Kraneveld

Our results, show that by targeting only 12 21-bps specific sequences, we are able to distinguish SARS-CoV-2, from any other virus (> 99%). In addition, with 85% accuracy is possible to predict if an infected person will need to be 345 hospitalized or will be asymptomatic. These findings could help to identify patients with SARS-CoV-2 that are susceptible to develop severe acute respiratory infection and make a better clinical management. Nevertheless, our conclusions hold only for the data currently at our disposal. Further testing is necessary to confirm these promising results so it is essential to create multidisciplinary 350 groups that work to stop the outbreak. Finally, as an interesting remark, by comparing the discovered sequences against other hosts, we noticed that from the 12 sequences exclusive to SARS-CoV-2, 1 of them appears in all of the 9 sequences from Manis Javanina. In contrast, 5 of the sequences of SARS-CoV-2 34 . CC-BY-NC 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.03.13.990242 doi: bioRxiv preprint appear in the only sample available from Rhinolophus Affinis. The addition of 355 these sequences sum up to 6 of the 12 sequences that we used to characterized the SARS-CoV-2. This is consistent with the findings of Zhang et al. [45] , and could point to the zootonic origin of the virus.