A data-driven and probabilistic approach for data integration, calibration, and analysis in mechanistic models of cellular processes

Abstract

Extracting biological knowledge from experimental data continues to be one of the main challenges toward a mechanistic understanding of cellular processes. Mathematical models can provide mechanistic insights into the dynamics of cellular processes but, their calibration requires quantitative measurements that cellular experiments typically cannot completely provide. Recent efforts to address this challenge have aimed to substitute quantitative measurements with readily available nonquantitative measurements with varying degrees of success. However, existing approaches for data type integration introduce ad hoc assumptions or encumber Bayesian model-calibration methods resulting in calibration biases that obscure model interpretation and compromise model accuracy and certainty. To address these limitations, this work introduces an unbiased data-driven and probabilistic methodology for quantitative and non- quantitative data integration in biological model calibration. The approach employs Bayesian parameter inference methodology to estimate the contributions of nonquantitative data to mechanistic model accuracy. The work reveals that far more nonquantitative measurements are needed to compensate for quantitative measurements, but these measurements can have diverse sources. The data-driven representation of measurement also provides potential insight into the mechanistic/dynamic predictors of nonquantitative phenotypes.