Machine learning, whole genome sequencing, and Mendelian randomization support a role of CRP on COVID-19 severity

Background The coronavirus disease 2019 (COVID-19) ranges from asymptomatic to very severe infection and death, largely depending on host factors, including genetics. We have investigated clinical and genetic data from 200 COVID-19 patients to search for factors predisposing to increased disease severity. Methods Patients were divided into non-hospitalized mild/pauci-symptomatic and hospitalized severe. An interpretable Machine Learning approach was applied to blood biomarkers while genome-wide associations were performed for COVID-19 severity. Finally, a possible causal role of chronic low-grade inflammation on COVID-19 severity was searched by Mendelian Randomization. Results A high severity predictive role was observed in our sample by Machine Learning for the C-Reactive Protein measured in the course of SARS-CoV-2 infection (iCRP). This was also suggested by evidence of association with variants known to be involved in the CRP levels in the general population (pCRP). Finally, a possible causal role of chronic low-grade inflammation on COVID-19 severity could be shown by Mendelian Randomization using publicly available summary statistics of two COVID-19 Genome-Wide Association Studies. Conclusions Consistent with previous results, a predictive role of CRP levels on COVID-19 severity was detected in our sample. Furthermore, Mendelian Randomization supported a causal role of genetically predicted chronic CRP levels.