We note, however, several limitations to our work. First and foremost, while we note that a few of our binding affinity predictions were borne out in experimentally validated SARS-CoV peptides (see Table S4 in the supplemental material), we acknowledge that ours was a study performed entirely in silico. As we are unable to obtain individual-level HLA typing and clinical outcome data for any real-world COVID-19 populations at this time, the data presented are theoretical in nature and are subject to many of the same limitations implicit in the MHC binding affinity prediction tool(s) upon which it is based. As such, we are unable to assess the relative importance of HLA type compared to known disease-modifying risk factors such as age and clinical comorbidities (4–9). We further note that peptide-MHC binding affinity is limited in its utility as a predictor of subsequent T-cell responses (54–56), and we did not study T-cell responses here. As such, we are ill-equipped to explore phenomena such as original antigenic sin (57–59), where prior exposure to a closely related infection(s) might trigger T-cell anergy (60–62) or immunopathogenesis (63) in the setting of a novel infection. We explored only a limited set of 145 well-studied HLA alleles but note that this analysis could be performed across a wider diversity of genotypes (48). Additionally, we did not assess genotypic heterogeneity or in vivo evolution of SARS-CoV-2, which could modify the repertoire of viral epitopes presented or could otherwise modulate virulence in an HLA-independent manner (64, 65) (https://nextstrain.org/ncov). We also did not address the potential for individual-level genetic variation in other proteins (e.g., angiotensin converting enzyme 2 [ACE2] or transmembrane serine protease 2 [TMPRSS2], essential host proteins for SARS-CoV-2 priming and cell entry [66]) to modulate the host-pathogen interface.