Fast and accurate computation of protein stability is increasingly important for e.g. protein engineering and protein misfolding diseases, but no consensus methods exist for important proteins such as globins, and performance may depend on the type of structural input given. This paper reports benchmarking of six protein stability calculators (POPMUSIC 2.1, I-Mutant 2.0, I-Mutant 3.0, CUPSAT, SDM, and mCSM) against 134 experimental stability changes for mutations of sperm-whale myoglobin. Six different high-resolution structures were used to test structure sensitivity that may impair protein calculations. The trend accuracy of the methods decreased as I-Mutant 2.0 (R=0.64-0.65), SDM (R=0.57-0.60), POPMUSIC2.1 (R=0.54-0.57), I-Mutant 3.0 (R=0.53-0.55), mCSM (R=0.35-0.47), and CUPSAT (R=0.25-0.48). The mean signed errors increased as SDM<CUPSAT<I-Mutant 2.0<I-Mutant 3.0<POPMUSIC 2.1<mCSM. Mean absolute errors increased as I-Mutant 2.0<I-Mutant 3.0<POPMUSIC 2.1<CUPSAT<SDM<mCSM. Structural sensitivity increased as I-Mutant 3.0 (0.05)<I-Mutant 2.0 (0.09)<POPMUSIC 2.1 (0.12)<SDM (0.18)<mCSM (0.27)<CUPSAT (0.58). Leaving out heterogeneous experimental data did not change conclusions. The distinct performances reveal room for improvement, but I-Mutant 2.0 is proficient for this purpose, as further validated against a data set of related cytochrome c like proteins. The results also emphasize the importance of high-quality crystal structures and reveal structure-dependent effects even in the near-atomic resolution limit.
- Mutant stability
- Protein calculation
- High-resolution crystal structure
- Structure sensitivity