Parameter sensitivity analysis of a human atrial cell model using multivariate regression

Atrial cell models form the building blocks of complex multicellular models and contain many input parameters within a large parameter space. A rapid systematic way of quantifying changes in model outputs due to input parameter variability would enhance mechanistic understanding. A parameter sensitivity study on input model conductances within the Courtemanche-Ramirez-Nattel (CRN) human atrial cell model was performed. Input maximal ionic conductances were varied by randomly scaling baseline model values, and used in single cell action potential (AP) simulations. Multivariable regression was performed to find regression matrix B which minimised differences between output Y and predicted Y * = XB. Regression values were mean-centred and normalized to SD, such that a +0.5 value implied parameter input 1SD above mean increased output by 0.5SD. AP determinants were compared (n = 100). dV=dt
_max
and V
_amp
showed strong sensitivity only to maximal sodium conductance G
_Na
, whilst APD and t
_peak
were weakly sensitive to multiple conductances. Predicted outputs showed strong correlation to measured values: 0.880 ≤ R
²
≤ 0.998 for measured AP determinants. Multivariable regression is a novel tool for examining parameter sensitivity of the CRN AP model and offers rapid insight to the relative importance of input parameters to specific outputs.