Machine Learning Informed Predictions of Nanoparticle Mobility and Fate in the Mucus Barrier

Diffusion and transport of nanomaterials through mucus, is of critical importance to many basic and applied areas of research such as drug delivery and infectious disease. However, it is often challenging to interpret the dynamics of nanoparticles within the mucus gel due to its inherently heterogeneous microstructure and biochemistry. In this study, we measured the diffusion of densely PEGylated nanoparticles (NP) in human airway mucus ex vivo using multiple particle tracking and utilized machine learning to classify NP movement as either traditional Brownian motion (BM) or one of two different models of anomalous diffusion, fractional Brownian motion (FBM) and continuous time random walk (CTRW). Specifically, we employed a physics-based neural network model to predict the modes of diffusion experienced by individual NP in human airway mucus. We observed rapidly diffusing NP primarily exhibit BM whereas CTRW and FBM exhibited lower diffusion rates. Given the use of muco-inert nanoparticles, the observed transition from diffusive (BM) to sub-diffusive (CTRW/FBM) motion is likely a result of patient-to-patient variation in mucus network pore size. Using mathematic models that account for the mode of NP diffusion, we predicted the percentage of nanoparticles that would cross the mucus barrier over time in human airway mucus with varied total solids concentration. We also applied this approach to explore the transport modes and predicted fate of influenza A virus within human mucus. These results provide new tools to evaluate the extent of synthetic and viral nanoparticle penetration through mucus in the lung and other tissues.