Machine Learning-Enhanced Solid-State Nanopore Analysis for Single-Molecule Discrimination of Albumin and Aptamer-Bound Glycated Albumin Complexes

Abstract

Discriminating glycated human serum albumin from unmodified albumin and aptamer-bound complexes at the single-molecule level is a sensing challenge unmet by conventional bulk analytical methods, which cannot resolve individual protein modification states. Here, we present a proof-of-concept solid-state nanopore platform combined with aptamer recognition and supervised machine learning for single-event discrimination of human serum albumin (HSA), glycated HSA (GHSA), and aptamer-bound GHSA (Apt+GHSA). Using a 14 nm silicon nitride nanopore, synthetic HSA, GHSA, and Apt+GHSA produced distinguishable ionic current blockade signatures, while dwell times remained similar. Serum-derived HSA showed broader, heterogeneous blockade distributions consistent with mixed albumin proteoforms; aptamer treatment generated a higher-blockade population consistent with Apt+GHSA-like events. A modified fluorescence aptamer assay estimated serum glycated albumin content at approximately 30%. To resolve overlapping signatures, five signal-derived features were extracted from individual translocation events and used to train supervised machine learning classifiers. Random Forest and Gradient Boosting achieved the highest three-class test accuracy of 94.50% for HSA, Apt+GHSA, and serum-derived HSA; Gradient Boosting achieved the highest macroaverage AUC of 0.9915. Applied to aptamer-treated serum-derived HSA, Gradient Boosting classified 54.6% of events as Apt+GHSA-like, reflecting aptamer-responsive nanopore signatures rather than a direct quantification of the glycated albumin percentage. This study establishes a nanopore–aptamer–machine learning framework for resolving albumin modification and binding states in biologically derived samples.