Fig. 4

Identification of IRGs.(A Deconvolution model based on support vector machine regression identifies IRGs. B Speaman correlation coefficients between predictors and their expression levels for all protein-coding genes in the TCGA HNSCC cohort. C Further screening obtained IRGs that were conserved in two HNSCC cohorts. D The difference of CRISPR- or RNAi-based gene dependency scores between IRGs and random Genes. E Differential expressed IRGs in two HNSCC cohorts (tumor vs normal, the most significant IRG is marked). F Venn diagram showing the shared differential expressed IRGs genes in two HNSCC cohorts. G Differential expressed Proteins in IRGs (tumor vs. normal, the most significant IRG is marked) and their correlation with gene expression in Chen’s cohort. H Hazard rations (Cox regression model) of IRGs in the TCGA cohort and their 95% confidence intervals (Top DEGs in all IRGs with significant prognostic relevance were labeled in the proportional hazards model)