Integrated transcriptomic analysis of COVID-19 stages and recovery: insights into key gene signatures, immune features, and diagnostic biomarkers through machine learning

Abstract

Background: COVID-19 progression and recovery involve complex gene expression changes and immune dysregulation, but their dynamic alterations remain poorly understood. Current clinical indicators lack precision in distinguishing severe cases, highlighting the need for molecular biomarkers and diagnostic tools. Methods: Three transcriptomic datasets were analyzed: 1) COVID-19 progression from Healthy, Moderate, Severe, to ICU patients; 2) recovery stages (1, 3, and 6 months) compared to Healthy controls; and 3) COVID-19 ICU versus non-ICU patients. Differential expression analysis, immune cell infiltration estimation, machine learning (LASSO regression and random forest), and functional enrichment were used to identify key genes and molecular mechanisms. Results: Gene expression analysis revealed dynamic changes during COVID-19 progression. Adaptive immune cells (e.g., B cells and T cells) decreased, while innate immune cells (e.g., monocytes and neutrophils) increased, particularly in ICU patients. Recovery analysis showed significantly reduced adaptive immune cells at 1 month, with partial recovery by 3 and 6 months. Machine learning identified CCR5, CYSLTR1, and KLRG1 as diagnostic biomarkers for distinguishing ICU from non-ICU patients, with AUC values of 0.916, 0.885, and 0.899, respectively. Conclusion: This study identified CCR5, CYSLTR1, and KLRG1 as efficient diagnostic biomarkers for severe COVID-19 using machine learning and revealed immune regulatory features across COVID-19 progression and recovery.