Quantum-Transport Informed Machine Learning for Identifying Tobacco-Induced Regioisomeric DNA Adducts.

Tobacco smoke contains a complex array of genotoxic carcinogens that form structurally diverse DNA adducts, driving mutagenesis and carcinogenesis. Among these, certain adducts exist as regioisomers, differing in the specific site of covalent attachment on the nucleobase, which in turn alters their structural and electronic properties. Detecting these adducts remains challenging due to subtle structural variations. To overcome the limitations of conventional protein-based nanopores, we developed a machine learning-empowered graphene nanogap platform integrating quantum transport analysis with a semisupervised framework. Distinct tunneling signatures extracted from transmission spectra and I-V characteristics serve as electronic fingerprints for precise adduct identification. Employing a self-training random forest classifier, the system achieved high accuracy in automated recognition. Our approach enables the rapid and real-time detection of tobacco carcinogen DNA adducts, advancing biomarker discovery and cancer risk assessment.