Fig. 1

Machine learning framework for large-scale identification and validation of HNSC cell states and ecosystems This schematic diagram illustrates the application of EcoTyper in HNSC patients. Initially, cell states were identified in a discovery cohort composed of single-cell sequencing data from multiple cohorts of primary HNSC patients. These cell states were subsequently validated using metastatic single-cell sequencing samples, and associations between cell state abundance and patient outcomes were analyzed. HNSC ecotypes were identified by examining co-occurrence patterns among cell states, and spatial transcriptomics was employed to validate the spatial distribution of both ecotypes and cell states. Finally, the association between tumor ecotypes and patient outcomes was examined, along with the analysis of the relationship between cell states, ecotypes, and treatment sensitivity across several immunotherapy and chemotherapy cohorts