Filtering evaluation of multiplex respiratory virus-like particles by machine learning-assisted surface-enhanced Raman spectroscopy

Abstract

The persistent threat of respiratory viral infections emphasizes the need for face masks with effective virus protection. Masks provide an immediate physical barrier against respiratory droplets, particularly without effective therapeutic interventions. Conventional filtration performance tests cannot replicate key viral characteristics or the real-world environment of viruses within respiratory droplets. Here, we developed a cutting-edge system for mask evaluation using machine learning (ML)-assisted surface-enhanced Raman spectroscopy (SERS) with multiplex respiratory virus-like particles (VLPs). Our microfluidic spray system generates human respiratory droplets containing VLPs with adjustable specific viral properties, simulating real-world transmission conditions. Additionally, the developed ML-based advanced one-dimensional convolutional neural network (1D-CNN) model efficiently analyzed complex SERS spectral datasets, quantifying the distribution of Raman dye-tagged VLPs with over 92 % accuracy. Consequently, this high-throughput, multiplexed system enables precise evaluation of mask filtration performance under realistic conditions and provides valuable insights into viral droplets across diverse environments, supporting evidence-based public health strategies to control respiratory infections.