Uncertainty-Aware Modeling in Integrated Epithelial Transport Networks

While computational mechanistic models can provide insight into the coupled transport of substrates through complex networks of membrane transporters and channels, commonly used deterministic models are often constrained by uncertainty in model structure and in parameters derived from sparse and limited data. As a result, models may lack generalizability, particularly when high confidence is required for testing mechanistic hypotheses. Here we present, as a proof-of-principle, a Bayesian framework for uncertainty-aware modeling that enables deeper mechanistic insight and more reliable hypothesis testing. As an illustrative example, we applied this framework to the rat choroid plexus epithelium, a tissue characterized by a coupled, nonlinear transport network for which quantitative data are limited. Using a thermodynamics-based model of coupled Na + , K + , Cl - , HCO 3 - , and water transport, we compared deterministic parameter assignment with Bayesian inference. The Bayesian framework yielded physiologically plausible parameter distributions for 14 cellular, transporter, and channel parameters, quantified predictive uncertainty and data informativeness, and enabled assessment of model structural adequacy. We show how the framework distinguishes between model-data mismatch arising from parameter uncertainty and that due to structural limitations, and identifies where additional data would reinforce predictive reliability. Predictions based on data-informed parameter ensembles reproduced expected stoichiometries (e.g., K⁺:Cl⁻ ≈ 1:2) and revealed testable mechanistic couplings, including constraints on NKCC1 flux imposed by Na⁺/K⁺-ATPase capacity, that are not accessible through nominal deterministic fits. Overall, this work provides a systematic approach for assessing, refining, and interpreting mechanistic models of epithelial transport under uncertainty that is intrinsic to biological systems.