New Nanopore-Based Method Combines Single-Molecule Detection and AI to Identify Proteins

Proteins carry out most functions in the human body, yet analyzing them—a necessity for understanding disease, developing drugs, and discovering biomarkers—remains highly complex. A team at the University of Geneva has now developed a method for identifying proteins molecule by molecule using a technology called nanopore detection, combined with artificial intelligence. The findings were published in the Journal of the American Chemical Society. 

Nanopore detection works by passing molecules through a tiny hole, just a few nanometers wide, embedded in a membrane. As each molecule passes through, it briefly disrupts an electrical current flowing through the pore, producing a characteristic change in that current— a kind of molecular fingerprint. By analyzing these signals, researchers can distinguish between different proteins, even highly similar ones.

A key challenge in applying nanopore technology to proteins is controlling their movement through the pore. “By nature, proteins carry complex electric charges and thus cannot be consistently controlled using electrophoretic forces alone, that is, the forces exerted by an electric field on charged molecules,” explained study leader Chan Cao. To address this, the team exploited a phenomenon called electro-osmotic flow—a liquid flow inside the nanopore that drives proteins through it regardless of their charge.

Because highly similar proteins can produce electrical signals that are difficult to distinguish, the researchers incorporated artificial intelligence to interpret the data. Each time a protein passes through the pore, it generates a complex signal. The team broke this signal down into measurable characteristics—such as duration and how the current changes over time—and fed them into an algorithm trained to associate patterns with specific proteins. By learning from known samples, the system can then recognize unknown proteins based on their fingerprint, even when differences are subtle.

The method enables single-molecule detection and label-free protein identification, removing the need for chemical tags or labels that can interfere with analysis. The team is now working to extend the approach further. “We are currently working on establishing a rational link between the measured electrical current and the protein sequence. This might make it possible not only to recognize proteins we have already measured, but also to directly analyze new, unknown protein samples,” said first author Verena Rukes.