AI-Guided SERS Defines a Pan-Cancer Diagnostic Biomarker.

Early and accurate detection of multiple cancers through a single test remains an unmet clinical need, hindered by current limitations in accuracy, throughput, automation, and multiplexing. Here, we present an AI-powered SERS chip that combines automated exosome capture with AI-enabled molecular fingerprinting to accurately distinguish ten common cancer types from a single serum test. The system employs a peptide-functionalized SERS chip enabling the selective enrichment of exosomes directly from patient serum, enhancing label-free Raman fingerprint signals. AI-driven spectral analysis achieved 97.4% accuracy in distinguishing cancer from healthy samples, 97.08% accuracy for early-stage cancer detection, and 93.89% accuracy in classifying ten common cancer types, including breast, thyroid, esophageal, kidney, pancreatic, duodenal, lung, colorectal, ovarian, and gastric cancers. Crucially, based on molecular profiling, we identified exosomal deoxyadenosine triphosphate as a promising pan-cancer biomarker consistently upregulated across diverse tumor types. This discovery establishes a potential pan-cancer diagnostic marker, while the fully automated, scalable platform offers significant promise for clinical translation in early and differential cancer diagnosis.