Machine Learning-Assisted High-Throughput Identification and Quantification of Protein Biomarkers with Printed Heterochains

Abstract

Advanced in vitro diagnosis technologies are highly desirable in early detection, prognosis, and progression monitoring of diseases. Here, we engineer a multiplex protein biosensing strategy based on the tunable liquid confinement self-assembly of multi-material heterochains, which show improved sensitivity, throughput, and accuracy compared to standard ELISA kits. By controlling the material combination and the number of ligand nanoparticles (NPs), we observe robust near-field enhancement as well as both strong electromagnetic resonance in polymer–semiconductor heterochains. In particular, their optical signals show a linear response to the coordination number of the semiconductor NPs in a wide range. Accordingly, a visible nanophotonic biosensor is developed by functionalizing antibodies on central polymer chains that can identify target proteins attached to semiconductor NPs. This allows for the specific detection of multiple protein biomarkers from healthy people and pancreatic cancer patients in one step with an ultralow detection limit (1 pg/mL). Furthermore, rapid and high-throughput quantification of protein expression levels in diverse clinical samples such as buffer, urine, and serum is achieved by combining a neural network algorithm, with an average accuracy of 97.3%. This work demonstrates that the heterochain-based biosensor is an exemplary candidate for constructing next-generation diagnostic tools and suitable for many clinical settings.