biostudies-literatureAkita RJapan Science and Technology Agency18607-18615https://www.ebi.ac.uk/biostudies/studies/S-EPMC12499889biostudies-literatureUnknown16(40)Solid-state nanopores are attracting attention as a label-free method for detecting diverse physical properties of biomarkers. However, improving molecular discrimination in complex biological samples remains a major challenge, partly due to uncertainty in selecting optimal measurement conditions. We developed a Voltage-Matrix Analysis that visualizes classification accuracy across multiple voltages using machine learning. We measured two tumor markers (CEA and CA15-3) individually and in mixtures using solid-state nanopores, applying Random Forest and Support Vector Machine classifiers. Overfitting occurred when baseline-involving features were used, necessitating optimization of the feature set, which led to voltage-independent high classification performance. For mixed samples, we estimated actual molecular ratios by combining classification probability histograms with detection frequency correction. We further tested mouse serum with and without centrifugation and achieved notable classification accuracy. These findings suggest that voltage-dependent structural changes influence molecular discrimination, and that our method may aid diagnosis of diseases lacking known biomarkers by identifying specific molecular population shifts.Chemical scienceVoltage-matrix nanopore profiling for the discrimination of protein mixtures.PMC12499889JPMJCR2231JPMJKP23H2Lysenko AIizuka RUemura SAkita RKawai DTsunoda TYokota TBoroevich KAfalseVoltage-matrix nanopore profiling for the discrimination of protein mixtures.Solid-state nanopores are attracting attention as a label-free method for detecting diverse physical properties of biomarkers. However, improving molecular discrimination in complex biological samples remains a major challenge, partly due to uncertainty in selecting optimal measurement conditions. We developed a Voltage-Matrix Analysis that visualizes classification accuracy across multiple voltages using machine learning. We measured two tumor markers (CEA and CA15-3) individually and in mixtures using solid-state nanopores, applying Random Forest and Support Vector Machine classifiers. Overfitting occurred when baseline-involving features were used, necessitating optimization of the feature set, which led to voltage-independent high classification performance. For mixed samples, we estimated actual molecular ratios by combining classification probability histograms with detection frequency correction. We further tested mouse serum with and without centrifugation and achieved notable classification accuracy. These findings suggest that voltage-dependent structural changes influence molecular discrimination, and that our method may aid diagnosis of diseases lacking known biomarkers by identifying specific molecular population shifts.2025-01-01T00:00:00Z2025 Oct2026-06-04T11:57:07.192Z2026-05-08T03:10:57.924ZS-EPMC124998894105910610.1039/d5sc05182g