Genelle F. Harrison, Wenyu Zhou, Meher Preethi Boorgula, Monica Campbell, Sameer Chavan, Brett R. Peterson, Bret Barnes, Rishi Porecha, Rasika A. Mathias, Alem Taye, Ivana Yang, Christopher Gignoux, Andrew Monte, Kathleen C. Barnes
The development of vaccinations, antivirals, and other interventions, has precipitated a transition from the acute phase of the SARS-COV-2 pandemic to an endemic infection cycle. In this phase of the pandemic, mitigating the worst outcomes of an infection with SARS-COV-2, and managing burdens on hospital systems, is reliant on our ability to identify persons at high risk of developing severe COVID-19 disease and administering treatments early and efficiently. Epigenomic pattern alterations in response to SARS-COV-2 infection are evident in circulating white blood cells. Early in the pandemic, we designed a machine learning platform that leveraged methylation risk scores (MRS) derived from differentially methylated signatures in infected and uninfected individuals (Konigsberg et al 2021 Comm. Med.). This approach yielded highly predictive scores, measured as a classification-threshold-invariant (AUC), for both presence of SARS-CoV-2 infection (AUC=93.6%) and as a measure for COVID-19 disease severity (AUC=79.1%-84.4%). The goal of this work was to determine the prediction efficacy of a previously developed sparse regression based MRS model towards infection status and disease severity. The population in consideration included vaccinated individuals who visited the Emergency Department after March 2021.
To accomplish this, we profiled peripheral blood samples from 304 additional patients (211 cases, 93 controls) on the customized Infinium MethylationEPIC array. These patients included a majority who had received at least one dose of the COVID-19 vaccine at the time of infection. Disease severity was assessed by hospitalization status, ICU admittance, administration of ventilator and death. This information along with vaccination status was extracted from patient electronic health records. We assessed the efficacy of our previously developed sparse regression based MRS model in predicting disease severity in vaccinated individuals. This was achieved by comparing infection status and severity AUCs between the aforementioned cohort and an unvaccinated cohort from the original study. In summary, our approach used a machine learning approach predicated on MRS to predict COVID-19 severity and demonstrated its utility in vaccinated and unvaccinated populations. Our study provides insights into how vaccination affects methylation status thereby offering protection from severe immune reactions due to deadly infections. In summary, our model previously designed and implemented on biospecimens from patients with SARS-CoV-2 infection prior to vaccination provides a powerful tool for clinicians to evaluate a patient’s propensity to develop severe COVID-19 disease regardless of vaccination status and to ensure early and appropriate interventions.
VIEW THE PUBLICATION
VIEW THE POSTER