Hybrid Physics-Informed and Bayesian Modeling of Single-Nanoparticle-Cell Adhesion Kinetics under Cytoskeletal...

Area of Science:

• Biophysics
• Nanomedicine
• Computational Biology

Background:

• Modeling nanoparticle-cell membrane interactions is crucial for nanomedicine.
• Traditional models struggle with biological variability and experimental noise.
• Existing frameworks are often limited by equilibrium-based assumptions.

Purpose of the Study:

• To develop an advanced hybrid Physics-Informed Neural Network (PINN) framework.
• To capture early-stage nanoparticle adhesion dynamics.
• To address limitations of traditional models in biological systems.

Main Methods:

• Integrated an analytical Standard Linear Solid model with a neural residual term.
• Employed an adaptive, uncertainty-aware loss weighting scheme.
• Utilized a physics-informed Bayesian framework with a heteroscedastic likelihood for parameter inference.
• Validated using single-nanoparticle force measurements on fibroblasts and cancer cells.

Main Results:

• The hybrid PINN model demonstrated condition-dependent predictive improvements over classical models.
• The model showed significant enhancements in predicting adhesion dynamics under cytoskeletal disruption.
• Achieved robust parameter inference for heterogeneous experimental data.
• Validated via leave-one-out cross-validation.

Conclusions:

• The hybrid model provides an interpretable approach for complex biophysical adhesion processes.
• Offers a generalizable framework for analyzing noisy, heterogeneous biological systems.
• Enables physically consistent parameter estimates across varying cytoskeletal perturbation levels.