Single-aminoacid discrimination in proteins with homogeneous nanopore sensors and neural networks

A technology capable of sequencing individual protein molecules would revolutionize our understanding biological processes. Nanopore can analyze single heteropolymer such as DNA by measuring the ionic current flowing through a nanometer hole made in an electrically insulating membrane. This is sensitive to monomer sequence. However, proteins are remarkably complex and identifying residue change remains challenge. In this work, I show that simple neural networks be trained recognize mutants. Although these quickly efficiently trained, their ability generalize independent experiment poor. Using thermal annealing protocol on nanopore sample, examining many mutants with same sensor measures aimed at reducing training data variability which produce increase generalizability network. approach, we obtain 100% correct assignment among 9 >50% experiments. Interestingly, network performance, compared random guess, improves more included dataset for discrimination. Engineered nanopores prepared high homogeneity coupled state-of-the-art analysis signals may enable single-molecule sequencing.