Fig. 4. Model robustness to train–validation–test cutoffs and noise.
a–c, The influence of train–validation–test cutoffs (a), shuffling (Synthetic_FoldX_ΔΔG_942723_shuffled) (b) and Gaussian noise (Synthetic_FoldX_ΔΔG_942723_gaussian_noise) (c) on model performance. Results are shown for 10-fold cross-validation in a and for a single fold, held-out test set in b and c. The Pearson’s correlation (r) values are shown in the scatter plots in a–c. For all histograms, the x axes were limited to −8 to +5 kcal mol−1 for clarity and the solid lines are kernel density estimates (KDEs). Pred, predicted; SDR, standard deviation ratio.