Machine Learning-Augmented Graphene Transistor Biosensing: Quantitative Platform Validation and Immunotesting of Hepatitis E

Abstract

Graphene-based chips face persistent sensor-tosensor variability due to manufacturing defects and polymer contamination, limiting their analytical reliability for healthcare applications. Here, we demonstrate that the integration of a machine learning (ML) model with graphene field-effect transistors (GFETs) enables quantitative and calibration-free analytical sensing. Using Random Forest Regression and field-effect-related figures of merit, the model enabled robust, quantitative predictions across analytes of varying chemical nature�from small ions to viral antigens. pH sensing was used as a reference system to validate the augmented platform. Compared with the reference analytical model, ML enabled a marked improvement of accuracy, from 93 to 97%, and a reduction of the coefficient of variability, from 14 to 3%. Then, the ML-integrated GFETs were applied to chloride detection, the gold standard for cystic fibrosis diagnosis. Finally, using GFETs functionalized with llama nanobodies, we targeted the ORF2 antigen of the Hepatitis E virus. ML integration significantly enhanced immunoassay sensitivity-specificity from 89−69% to 100−100% and allowed the quantitative prediction of antigen concentration. Furthermore, the ML-augmented test demonstrated a strong performance for HEV antigen detection in capillary blood samples without the need for any sample pretreatment.