Molecular Classification of Patients With COVID‐19 Based on Transcriptional Profiling

Background COVID‐19 has caused over 7 million deaths worldwide and remains a critical public health threat. The marked heterogeneity in immune responses among patients poses challenges for targeted treatment. Molecular classification is essential for guiding precision therapies. Methods We performed unsupervised consensus clustering on blood transcriptomic data from 351 COVID‐19 patients to identify molecular endotypes and validated the classification in an independent cohort of 56 patients. To identify robust endotype‐specific biomarkers, we applied XGBoost, LASSO, and random forest algorithms. Results Three endotypes with distinct biological and clinical profiles were identified. Endotype 1, associated with favorable outcomes, showed enriched DNA replication pathways and elevated IL7 expression. Endotype 2 featured hypoxia and angiotensin‐related pathways. Endotype 3 exhibited TLR4 activation, IL‐1β upregulation, and impaired NK cytotoxicity, correlating with poor outcomes. All endotypes shared type I interferon activation. Predictive biomarker pairs included STAT4:S100A11 (endotype 1), SLC4A1:RPL31 (endotype 2), and RALB:MTR (endotype 3), enabling endotype classification with high accuracy. Importantly, these biomarker genes can be reliably measured in peripheral blood using RT‐qPCR, making the classification model feasible for clinical application. Conclusions This molecular classification reveals heterogeneity in COVID‐19 and proposes biomarker‐guided strategies for patient stratification and management.