Early diagnosis and molecular classification of gastric cancer peritoneal metastasis by deep learning-driven single-cell lipidomics

Peritoneal metastasis is the most lethal manifestation of gastric cancer (GC), as conventional diagnostic methods cannot reliably detect occult free GC cells in the peritoneal cavity at an early stage. Here, we present a workflow for early detection and molecular classification of free GC cells using label-free metabolite-based mass cytometry for single-cell lipidomic profiling integrated with deep learning. We obtain 41,627 single-cell lipidomic profiles from 92 samples in total and identify a distinct lipidomic fingerprint that robustly discriminates between free GC cells and benign peritoneal cells. To further interpret complex lipidomic profiles, we develop a deep learning-based graph neural network classifier to identify rare free GC cells within peritoneal ecosystem. Beyond diagnostic discrimination, we propose a clinically relevant metabolic classification based on HER2 subtype-specific lipidomic heterogeneity. Our study advances the early diagnosis and molecular classification of peritoneal metastasis in GC, offering potential for metabolism-guided intervention.