Integrating structurally defined DNA-carbon nanotube sensors with machine learning for cancer detection.

Abstract

Liquid biopsy is a promising, noninvasive approach for cancer detection, but current methods often trade off accuracy, operability, and cost. To address these limitations, we introduce an artificial perception system (APS) for liquid biopsy that combines a structurally defined DNA-carbon nanotube sensor array with machine learning (ML) models. The array produces multichannel fluorescence fingerprints from serum, which are decoded by ML models to classify disease state. In a total of 253 serum samples spanning liver, lung, and ovarian cancers and noncancer controls, the APS achieved mean sensitivity of 89% and specificity of 96%. Notably, early-stage lung cancer was detected with 92% sensitivity and 95% specificity at an estimated cost of ∼$4 USD per test. Insights from SHAP analysis and Mantel test revealed the detection mechanisms of APS, supporting biological plausibility and clinical translation. These results highlight a path toward accurate, scalable, and affordable multicancer detection and early cancer screening.