Machine learning and single-cell approaches identify an NAD metabolism-related signature for predicting immunotherapy outcomes

To investigate the role of aging-associated nicotinamide adenine dinucleotide (NAD) metabolic reprogramming in cancer progression and immunotherapy, we constructed a comprehensive multi-omics framework employing machine learning and single-cell analysis. We collected 13 NAD metabolism-related pathways comprising 130 NAD metabolism-derived genes (NMDGs) and found significant correlations between aging and NMDGs at the genomic, epigenomic, and transcriptomic levels. Through Multiscale Embedded Gene Co-expression Network Analysis (MEGENA), we identified 564 modules significantly enriched in NAD metabolism pathways, with 153 (27%) associated with aging. We developed the NADLnc computational framework to systematically identify long non-coding RNAs associated with NAD-related pathways, identifying approximately 270 lncRNAs per cancer type, with significant associations between NAD metabolism-related lncRNAs and aging across 8 cancer types. Based on 32 NAD regulators, we developed NADscore, which effectively stratifies 20 cancer types, including bladder urothelial carcinoma, and correlates with poor prognosis. We constructed an XGBoost machine learning model (NADm) containing 5 NMDGs (NAMPT, GOT1, TPI1, MDH2, and NUDT12) that demonstrated robust predictive performance for melanoma immunotherapy with AUC values of 0.906 and 0.851, and cross-cancer applicability in kidney renal clear cell carcinoma (AUC = 0.7905) and urothelial carcinoma (AUC = 0.7569). Through propensity score matching, we identified multidimensional molecular features associated with NAD dysregulation, including 5,044 mRNA, 104 protein, 298 miRNA, and 4,062 methylation changes, and identified 34 potential therapeutic drugs targeting NAD dysregulation. Furthermore, single-cell and spatial transcriptome analyses revealed aging-related NAD dysregulation predominantly in the tumor center and surrounding macrophages. This framework provides promising avenues for incorporating NAD metabolism-based strategies into precision oncology, with extensive resources available at https://github.com/NoviZhou/NAD .