Fitting two human atrial cell models to experimental data using Bayesian history matching | JoVE Visualize

Area of Science:

Computational biology
Pharmacology
Biophysics

Background:

Cardiac cell models are crucial for safety pharmacology but difficult to calibrate due to numerous parameters and biological variability.
Action potentials exhibit beat-to-beat and cell-to-cell variations, complicating model parameterization.
High-dimensional parameter spaces make traditional calibration methods challenging.

Purpose of the Study:

To demonstrate the efficacy of history matching for calibrating detailed human atrial action potential models.
To refine model parameters, specifically ion channel and exchanger maximum conductances, against experimental biomarkers.
To explore the parameter space and identify parameter covariation consistent with experimental data.

Main Methods:

Utilized history matching, a technique employing Gaussian process emulators of computational models.
Constructed emulators from a limited number of model runs (approx. 10^2) and executed them millions of times (>10^6) computationally efficiently.
Quantified implausibility by comparing emulator outputs to experimental biomarkers, incorporating both experimental and emulator variance.
Iteratively reduced the non-implausible parameter space through repeated application of the history matching process.

Main Results:

Successfully calibrated two detailed human atrial cardiac cell models against experimental action potential biomarker measurements.
Generated sets of parameters that allow for the simulation of variable action potentials.
Revealed that model parameters do not converge to narrow ranges but can co-vary widely while remaining consistent with biomarkers.
Identified correlations between certain biomarkers, suggesting a need for improved action potential descriptors.

Conclusions:

History matching is an effective method for calibrating complex cardiac cell models against experimental data.
The approach facilitates the generation of parameter sets that reproduce observed biological variability in action potentials.
The study highlights the potential for parameter co-variation and suggests the need for enhanced biomarkers to fully characterize action potential dynamics.