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

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