Fig. 4. Scatter plots of AΔGDT-HA as a function of start GDT-HA score for best of top 5 models from GSFE-refinement for 3DRobot dataset. Corresponding LR/λ/W combination is noted on top of each plot. Effects of variation are exhibited for (A–C) learning rates; (D–F) structural restraints and (G–I) weights for approximating local priors. More scattered plots of ΔRMSD and ΔGDT-HA are available in ESI (Fig. S4–S6).†.