Single-molecule techniques for protein sequencing are making headway towards single-cell proteomics and are projected to propel our understanding of cellular biology and disease. Yet, single cell proteomics presents a substantial unmet challenge due to the unavailability of protein amplification techniques, and the vast dynamic-range of protein expression in cells. Here, we describe and computationally investigate the feasibility of a novel approach for single-protein identification using tri-color fluorescence and plasmonic-nanopore devices. Comprehensive computer simulations of denatured protein translocation processes through the nanopores show that the tri-color fluorescence time-traces retain sufficient information to permit pattern-recognition algorithms to correctly identify the vast majority of proteins in the human proteome. Importantly, even when taking into account realistic experimental conditions, which restrict the spatial and temporal resolutions as well as the labeling efficiency, and add substantial noise, a deep-learning protein classifier achieves 97% whole-proteome accuracies. Applying our approach for protein datasets of clinical relevancy, such as the plasma proteome or cytokine panels, we obtain ~98% correct protein identification. This study suggests the feasibility of a method for accurate and high-throughput protein identification, which is highly versatile and applicable.
Macromolecules identification methods are central for most biological and biomedical studies, and while the field of genomics advanced to single-molecule resolution, the proteomic field still relies on bulk and costly techniques. We describe a solution for single protein identification, based on the analysis of optical traces obtained from fluorescently-labeled proteins threaded through a nanopore and processed by a pattern recognition algorithm. To evaluate the feasibility of our method we constructed computer simulations of the system, producing and analyzing nearly 10 8 individual protein translocations from the human Swiss-Prot database. Our results suggest protein identification of >95% for the whole human proteome, even under non-ideal conditions. These results constitute the basis for a novel whole proteome identification method, with single molecule resolution.