Prediction efficiency of atom models – prediction accuracy for predicting thermal ΔΔG values. Structural environment specific comparison of 5 different atom classification models used in the prediction of changes in protein stability. Correlation (1) and prediction accuracy (2) for predicting thermal ΔΔG values. Prediction efficiencies in 11 different structural regions were compared using CL11 classification method . This method classifies mutations into 11 structural regions [helices (0–2, 2–30, 30–60, 60+), strands (0–5, 5–35, 35+) and others (0–10, 10–42, 42–67, 67+)]. MF40, SA28 and LN24 stand for the Melo-Feytmans, SATIS and Li-Nussinov atom models respectively.