High throughput analysis of rare nanoparticles with deep-enhanced sensitivity via unsupervised denoising
Yuichiro Iwamoto, Benjamin Salmon, Yusuke Yoshioka, Ryosuke Kojima, Alexander Krull, Sadao Ota
Abstract
The large-scale multiparametric analysis of individual nanoparticles is increasingly vital in the diverse fields of biology, medicine, and materials science. However, the current methods struggle with the tradeoff between measurement scalability and sensitivity, especially when identifying rare nanoparticles in heterogeneous mixtures. By developing and combining an unsupervised deep learning-based denoising method and an optofluidic device tuned for nanoparticle detection, we realize a nanoparticle analyzer that simultaneously achieves high scalability, throughput, and sensitivity levels; we name this approach “Deep Nanometry” (DNM). DNM detects polystyrene beads with a detection of limit of 30 nm at a throughput of over 100,000 events/second. The sensitive and scalable DNM directly detects rare target extracellular vesicles (EVs) in non-purified serum, making up as little as 0.002% of the 1,214,392 total particles. Moreover, DNM accurately and sufficiently counts diagnostic marker EVs present in only 0.93% and 0.17% of particle detections in sera of colorectal cancer patients and healthy controls, demonstrating its potential application to the early detection of colorectal cancer. Rare cell-derived vesicles have the potential to play a key role in early disease detection. Here, authors introduce a highly sensitive and high throughput nanoparticle analyzer, which combines unsupervised deep learning denoising with specialized hardware to detect rare disease-related vesicles.