In Situ Probing of Polymer Residence on the Nanoparticle Surface by Combining Deep Learning and Single-Molecule Fluorescence Tracking.

This study presents a method for in situ analysis of the adsorption and desorption of polymers on the nanoparticle surface within entangled polymer solutions. This method is based on the principle that when a polymer adsorbs onto the nanoparticle surface, the diffusion coefficient of the polymer becomes equivalent to that of the nanoparticle. Consequently, adsorption events can be identified by properly detecting diffusion-state transitions in trajectories acquired through single-molecule fluorescence tracking experiments. This method, which involves numerically generated trajectories, training of a 1D convolutional neural network (1D-CNN), and state prediction, was validated experimentally. When the nanoparticle-to-polymer diffusion coefficient ratio is as high as 0.5, the trained 1D-CNN model can still achieve 85% accuracy and an F1 score of 0.86 in adsorption-state identification. These results significantly outperform those of the hidden Markov model and the threshold-based method. Using the proposed method, we found that in aqueous solutions containing entangled poly(ethylene glycol) (PEG) chains and silica nanoparticles of varying sizes, the adsorption probability of PEGs onto the nanoparticle surface decreases, while the adsorption duration increases as the nanoparticle size decreases.