Figure 4

The root mean square error between predicted and measured ΔΔ G values in the validation sets (σ _c ) is given as a function of ΔΔ G _P (continuous line) and ΔΔ G _M (dashed line). The dataset of mutations was divided in non-overlapping subsets corresponding to intervals of ΔΔG_P (or ΔΔG_M) of 0.25 kcal/mol width, and σ_c was computed on each subset separately using eq. 4. Subsets containing less than 10 mutations were dismissed.