Integration of single-cell regulon atlas and bulk RNA-seq for individualized prognostic prediction in stomach adenocarcinoma.

Transcriptional regulators reflect cellular heterogeneity and are key for prognostic modeling. Given the poor prognosis of stomach adenocarcinoma (STAD), regulator-derived signatures are vital for risk stratification. Using multi-stage scRNA-seq data, we delineated the transcriptional regulatory landscape of STAD and identified Helicobacter pylori-associated epithelial heterogeneity in intestinal metaplasia. We then developed a 23-regulator machine learning-based STAD prognostic signature (SPS) from malignant epithelial cells to predict overall survival (OS). Patients with high-SPS exhibited significantly worse OS than patients with low-SPS (hazard ratio [HR] = 1.50, 95% CI: 1.09-2.09, log rank p = 7.11 × 10-3). Notably, SPS outperformed other established STAD prognosis models across various independent datasets and remained an independent prognostic factor after adjusting for clinical and pathologic factors. Moreover, integrating SPS with tumor stage and age showed superior accuracy to stage alone. Collectively, our study establishes a robust regulator-based prognostic signature, holding potential to facilitate precision prognostication in STAD.