Dissecting the statistical properties of the Linear Extrapolation Method of determining protein stability
When protein stability is measured by denaturant induced unfolding the linear extrapolation method is usually used to analyse the data. This method is based on the observation that the change in Gibbs free energy associated with unfolding, $Δ_rG$, is often found to be a linear function of the denaturant concentration, $D$. The free energy change of unfolding in the absence of denaturant, $Δ_rG_0$, is estimated by extrapolation from this linear relationship. Data analysis is generally done by nonlinear least-squares regression to obtain estimates of the parameters as well as confidence intervals. We have compared different methods for calculating confidence intervals of the parameters and found that a simple method based on linear theory gives as good, if not better, results than more advanced methods. We have also compared three different parameterizations of the linear extrapolation method and show that one of the forms, $Δ_rG(D) = Δ_rG_0 - mD$, is problematic since the value of $Δ_rG_0$ and that of the $m$-value are correlated in the nonlinear least-squares analysis. Parameter correlation can in some cases cause problems in the estimation of confidence-intervals and -regions and should be avoided when possible. Two alternative parameterizations, $Δ_rG(D) = -m(D-D_{50})$ and $Δ_rG(D) = Δ_rG_0(1-D/D_{50})$, where $D_{50}$ is the midpoint of the transition region show much less correlation between parameters.