This benchmark tests the ability of computational alanine scanning protocols to recapitulate the results of analogous experimental alanine scanning results. Like ΔΔG, these protocols predict the energetic effect on folding or binding of a point mutation. The benchmark set consists of a previously published set of 233 mutations from experimental alanine scanning experiments in 19 different protein-protein interfaces with known crystal structures (see Kortemme & Baker, 2002 below).
As in ΔΔG, the metrics used to measure success in this benchmark are: i) the linear correlation (Pearson coefficient) between experimental and predicted values; ii) the mean absolute error (MAE) of same; and iii) the stability classification accuracy, which measures whether a mutation was correctly predicted to be stabilizing, destabilizing, or neutral.
We have re-implemented a previously published alanine scanning protocol in the current version of Rosetta in order to determine the current performance of this method. Unlike the generalized ΔΔG protocol, which minimizes the entire protein structure, the alanine scanning protocol avoids any perturbation of the backbone or side chains other than the residue being mutated, which is replaced with alanine.
The ΔΔG of binding upon mutation to alanine is calculated using the following equation, in which Rosetta total energy is used to estimate the ΔG of folding of each of the six terms:
ΔΔGbind = (ΔGMUTcomplex - ΔGMUTpartner A - ΔGMUTpartner B) - (ΔGWTcomplex - ΔGWTpartner A - ΔGWTpartner B)