Deciphering Host Genomic Pathway Landscapes in Severe COVID-19 by Exome Sequencing of a Large Admixed Brazilian Cohort.

The host genetic factors that resulted in the heterogeneous clinical outcomes of COVID-19 remained poorly understood, especially in admixed populations that are underrepresented in global genomic studies. In this study, we investigated the genetic landscape associated with COVID-19 severity using whole-exome sequencing data from 325 Brazilian individuals recruited between 2020 and 2023. To prioritize variants related to disease severity, we applied a machine learning framework combining Support Vector Machine - Recursive Feature Elimination and SHAP interpretation. This approach encompasses 25 variants that contributed to the model's discrimination between severe and non-severe cases (AUC = 0.94) in our cohort. Subsequently, logistic regression analyses adjusted for age, sex, principal components of ancestry, and significant comorbidities confirmed several loci associated with severity after correction for multiple testing (FDR < 0.05). The prioritized genes were grouped into biological axes, including mitochondrial bioenergetics, neuroimmune signaling, complement activation, and epithelial structural integrity. Variants in genes such as MUC16, SLC4A2, LAPS1, ARRDC2, and CCSER1 showed associations with disease severity. Furthermore, functional annotation using CADD and regulatory exploration using GTEx suggested that several loci may influence gene regulation in lung, immune, and neural tissues. Instead of implicating single causal variants, our findings highlight pathway-level genetic architectures that may contribute to the variability of the host response to COVID-19. These findings provide exploratory insights into potential host genetic pathways that may be involved in disease variability in admixed environments. However, replication in independent cohorts and functional validation are still needed.