Label-Free Imaging of Single Proteins and Binding Dynamics via Deep Learning-Enhanced Plasmonic Scattering Microscopy

Determining protein binding is fundamental to deciphering biochemical mechanisms and engineering advanced biosensors, yet label-free imaging of single-protein binding dynamics remains challenging. Here, we introduce plasmonic scattering microscopy integrated with a spatiotemporal deep-learning framework that continuously isolates and tracks single unlabeled proteins from complex backgrounds. By leveraging a tailored recurrent neural network, our approach achieves high-throughput, automatic, label-free tracking of single proteins. By enabling label-free imaging and real-time trajectory analysis, this system directly discriminates between transient and stable binding events via residence time measurements, resolves nanoscale protein motions, and quantifies the binding thermodynamics. This multifaceted analysis establishes a quantitative framework to distinguish specific from nonspecific interactions, which is a long-standing challenge in biosensing. We anticipate that this method will establish new benchmarks for detecting specific protein binding in low-abundance immunosensing and extend label-free single-protein analysis to increasingly complex environments, such as biofluids.