| ============================================================================================================== | |
| Dataset: ic50 — Control vs competitors (NB-corrected t on outer folds; Holm across competitors) | |
| ============================================================================================================== | |
| Control exp_id: polyatomic_polyatomic | |
| k folds: 5, alpha: 0.05 | |
| Model (exp_id) | Test RMSE (95% CI) | Test MAE (95% CI) | Val RMSE mean±sd | Val MAE mean±sd | |
| ----------------------------------------------------------------------------------------------------------------------------------------------- | |
| gat_ecfp | 0.756567 [0.685188, 0.831186] | 0.588957 [0.543713, 0.638853] | 0.768333 ± 0.038465 | 0.595889 ± 0.016349 | |
| gat_selfies | 0.739957 [0.677242, 0.811344] | 0.588332 [0.544694, 0.632072] | 0.780903 ± 0.049973 | 0.608905 ± 0.019732 | |
| gat_smiles | 0.741486 [0.672216, 0.814339] | 0.580751 [0.535140, 0.625264] | 0.781717 ± 0.049394 | 0.608372 ± 0.018920 | |
| gcn_ecfp | 0.771808 [0.698955, 0.846373] | 0.595372 [0.545556, 0.647247] | 0.763733 ± 0.036233 | 0.596118 ± 0.018796 | |
| gcn_selfies | 0.745231 [0.676725, 0.815795] | 0.588968 [0.546651, 0.634146] | 0.782497 ± 0.051083 | 0.606765 ± 0.021751 | |
| gcn_smiles | 0.740936 [0.671876, 0.811042] | 0.586459 [0.545321, 0.631808] | 0.782171 ± 0.050774 | 0.607169 ± 0.020282 | |
| gin_ecfp | 0.764419 [0.687659, 0.835018] | 0.592957 [0.544833, 0.642144] | 0.782464 ± 0.035356 | 0.604945 ± 0.013472 | |
| gin_selfies | 0.740795 [0.676709, 0.812728] | 0.589688 [0.545839, 0.635794] | 0.783037 ± 0.050806 | 0.610020 ± 0.019629 | |
| gin_smiles | 0.739775 [0.670553, 0.817909] | 0.586204 [0.541501, 0.630134] | 0.783117 ± 0.051127 | 0.608843 ± 0.020055 | |
| polyatomic_polyatomic | 0.749880 [0.688620, 0.817071] | 0.606406 [0.565874, 0.652223] | 0.756392 ± 0.037301 | 0.596187 ± 0.018479 | |
| sage_ecfp | 0.784046 [0.705414, 0.860384] | 0.603648 [0.556301, 0.652519] | 0.763812 ± 0.036209 | 0.591910 ± 0.014446 | |
| sage_selfies | 0.735632 [0.671116, 0.810552] | 0.584244 [0.540345, 0.630018] | 0.782491 ± 0.051402 | 0.609728 ± 0.021232 | |
| sage_smiles | 0.741967 [0.672369, 0.819359] | 0.581619 [0.539225, 0.628283] | 0.781917 ± 0.050153 | 0.608085 ± 0.019509 | |
| --- NB-corrected t (outer folds) per competitor --- | |
| comparison mean_diff_RMSE(comp-ctrl) t_NB_RMSE p_one_sided_RMSE mean_diff_MAE(comp-ctrl) t_NB_MAE p_one_sided_MAE NB_CI_RMSE_low NB_CI_RMSE_high NB_CI_MAE_low NB_CI_MAE_high | |
| polyatomic_polyatomic vs gat_ecfp 0.011941 1.098923 0.166751 -0.000297 -0.022502 0.508437 -0.018228 0.042110 -0.036964 0.036370 | |
| polyatomic_polyatomic vs gat_selfies 0.024511 1.203391 0.147587 0.012719 1.178996 0.151870 -0.032041 0.081063 -0.017233 0.042670 | |
| polyatomic_polyatomic vs gat_smiles 0.025325 1.237717 0.141754 0.012186 1.158887 0.155487 -0.031484 0.082134 -0.017009 0.041380 | |
| polyatomic_polyatomic vs gcn_ecfp 0.007341 0.692586 0.263339 -0.000069 -0.008186 0.503070 -0.022087 0.036769 -0.023357 0.023219 | |
| polyatomic_polyatomic vs gcn_selfies 0.026105 1.276520 0.135423 0.010578 0.840306 0.224016 -0.030674 0.082884 -0.024373 0.045529 | |
| polyatomic_polyatomic vs gcn_smiles 0.025779 1.282266 0.134509 0.010983 1.028327 0.180956 -0.030040 0.081599 -0.018670 0.040636 | |
| polyatomic_polyatomic vs gin_ecfp 0.026072 5.608322 0.002483 0.008759 0.824569 0.227974 0.013165 0.038979 -0.020733 0.038250 | |
| polyatomic_polyatomic vs gin_selfies 0.026646 1.373598 0.120761 0.013833 1.507133 0.103127 -0.027213 0.080504 -0.011650 0.039316 | |
| polyatomic_polyatomic vs gin_smiles 0.026725 1.338463 0.125879 0.012657 1.295786 0.132382 -0.028712 0.082162 -0.014462 0.039775 | |
| polyatomic_polyatomic vs sage_ecfp 0.007421 0.788797 0.237176 -0.004276 -0.463159 0.666336 -0.018699 0.033541 -0.029911 0.021359 | |
| polyatomic_polyatomic vs sage_selfies 0.026099 1.238986 0.141542 0.013542 1.205766 0.147177 -0.032386 0.084584 -0.017640 0.044723 | |
| polyatomic_polyatomic vs sage_smiles 0.025525 1.244014 0.140707 0.011899 1.050054 0.176473 -0.031443 0.082493 -0.019562 0.043360 | |
| --- Holm-adjusted p-values (RMSE family) --- | |
| comparison p_raw p_holm Significant | |
| polyatomic_polyatomic vs gin_ecfp 0.002483 0.029792 True | |
| polyatomic_polyatomic vs gin_selfies 0.120761 1.000000 False | |
| polyatomic_polyatomic vs gin_smiles 0.125879 1.000000 False | |
| polyatomic_polyatomic vs gcn_smiles 0.134509 1.000000 False | |
| polyatomic_polyatomic vs gcn_selfies 0.135423 1.000000 False | |
| polyatomic_polyatomic vs sage_smiles 0.140707 1.000000 False | |
| polyatomic_polyatomic vs sage_selfies 0.141542 1.000000 False | |
| polyatomic_polyatomic vs gat_smiles 0.141754 1.000000 False | |
| polyatomic_polyatomic vs gat_selfies 0.147587 1.000000 False | |
| polyatomic_polyatomic vs gat_ecfp 0.166751 1.000000 False | |
| polyatomic_polyatomic vs sage_ecfp 0.237176 1.000000 False | |
| polyatomic_polyatomic vs gcn_ecfp 0.263339 1.000000 False | |
| --- Holm-adjusted p-values (MAE family) --- | |
| comparison p_raw p_holm Significant | |
| polyatomic_polyatomic vs gin_selfies 0.103127 1.000000 False | |
| polyatomic_polyatomic vs gin_smiles 0.132382 1.000000 False | |
| polyatomic_polyatomic vs sage_selfies 0.147177 1.000000 False | |
| polyatomic_polyatomic vs gat_selfies 0.151870 1.000000 False | |
| polyatomic_polyatomic vs gat_smiles 0.155487 1.000000 False | |
| polyatomic_polyatomic vs sage_smiles 0.176473 1.000000 False | |
| polyatomic_polyatomic vs gcn_smiles 0.180956 1.000000 False | |
| polyatomic_polyatomic vs gcn_selfies 0.224016 1.000000 False | |
| polyatomic_polyatomic vs gin_ecfp 0.227974 1.000000 False | |
| polyatomic_polyatomic vs gcn_ecfp 0.503070 1.000000 False | |
| polyatomic_polyatomic vs gat_ecfp 0.508437 1.000000 False | |
| polyatomic_polyatomic vs sage_ecfp 0.666336 1.000000 False | |
| ============================================================================================================== | |
| Notes: | |
| • Tests are within-dataset, one-sided for control superiority, on outer-fold differences with Nadeau–Bengio SE correction (df = k-1). | |
| • Holm controls family-wise error across competitors per metric family. | |
| • Held-out Test metrics above are for context only; no fold-based omnibus tests are used. | |