Full-Length Protein Classification through Sequence Reading by Solid-State Subnanopores

Nanopore sequencing for full-length proteins is daunting due to inaccuracy of fingerprinting amino acids with ionic current and lack of theoretical models. Here, we demonstrate that nanopores drilled in 5 nm-thick membranes with constriction diameter <1 nm, i.e., subnanopores, can site-specifically read the primary structure of amyloid-β (Aβ) isoforms after 50-fold coverage. Molecular dynamics simulations reveal that charged side chains near the constriction swing and modulate the volume exclusion of neighboring residues, resulting in a sequence context-dependent signal altered between translocation orientations. Orientation of individual label-free molecules is inferred by clustering, which yields high-fidelity empirical consensus that strongly correlates with computational trajectories (PCC = 0.78 ± 0.03). A homemade self-supervised learning method categorizes Aβ42 and Aβ40 at the single-molecule level (AUC = 0.88 ± 0.02) and predicts the mole fraction in agreement with actual ratios. Solid-state subnanopores offer a straightforward strategy for proteoform single-molecule classification and sequence extraction in mixtures.