t-REML for robust heteroscedastic regression analysis of mitochondrial power

The paper reports a long series of observations on two statistically independent variates which have different nonlinear regressions on the same parameter vector. In our analysis of the data, heteroscedasticity is handled by residual maximum likelihood, REML, but to cope with outliers as well, REML based on a t-specification, called t-REML, is introduced. The data are the results of factorial experiments designed to model the quantitative performance of mitochondria generating aerobic power. The response variates are flux, measured as respiration, and ATP potential as log(ATP/ADP). The background of metabolism and oxidative phosphorylation is briefly sketched.