Disease-associated NK-cell state remodeling reveals transcriptional profiles predictive of clinical phenotypes

NKceiver trained on Human NK-cell Atlas profiles from over 1 million NK cells distinguishes 18 diseases and predicts COVID-19 severity and immune checkpoint inhibitor response with retained performance in unseen patients.

• Why it matters: NK-cell plasticity across pathological contexts and its link to patient-level disease phenotypes is poorly defined, limiting interpretable biomarkers for clinical stratification.
• What they did: The Human NK-cell Atlas (HNKA) was built from 1,632 donors spanning 50 diseases, mapping a shared NK-cell state architecture and identifying disease-associated state remodeling with state-specific transcriptional profiles.
• The result: Disease-biased redistribution and transcriptional displacement localized remodeling to specific NK-cell states (e.g., CD56dim NK ISG15 and CD56bright NK ITM2C), and cytokine perturbations recapitulated key features, enabling patient-level prediction.