Machine Learning and Enrichment Analysis Reveal Persistent Protein Signatures in Longitudinal Post-COVID Symptom Trajectories

Résumé fourni par la source

Post-COVID syndrome (PCS) is characterised by persistent symptoms affecting multiple organ systems. Despite evidence of immune dysregulation, the protein signatures driving PCS remain unclear. This study applies stratified machine learning and enrichment analysis to identify protein markers distinguishing PCS groups from recovered individuals longitudinally. We analysed samples from 365 individuals who had been hospitalised for COVID-19, classifying them into four symptom-based groups (Fatigue, Affective, Cardiopulmonary, and Gastrointestinal) and a recovered group. Up to 384 inflammatory plasma proteins were measured at two research visits (5 and 12 months post-hospital discharge) using Olink Explore. Stratified, cross-validated penalised logistic regression models were trained and evaluated using AUC, accuracy, and F1-score. Enrichment analysis was performed to identify functionally related pathways (5% FDR cutoff). Overall, several biomarkers (CCL7, CTRC, FCAR, IL10RA, LAMP3, MVK, TPSAB1, LGALS9) were consistently upregulated across most groups, suggesting a shared systemic immune dysregulation mechanism. Per-group analysis revealed additional insights: Fatigue (HSD11B1, CKMT1A, MATN2), Gastrointestinal (CLEC4C, CKMT1A, CNTNAP2), and Affective (MATN2, ISM1). Additionally, enrichment analysis identified three key pathways underlying PCS: Persistent Immune Activation, Interferon & Antimicrobial Response Dysregulation, and Endothelial Dysfunction & Tissue Homeostasis Disruption. Our findings suggest that all PCS groups share immune and vascular dysregulation as a core mechanism, with some symptom-specific proteins identified for each group.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.