_tier2_early_stop.log

/home/sam/anaconda3/lib/python3.13/site-packages/muon/_core/preproc.py:31: FutureWarning: `__version__` is deprecated, use `importlib.metadata.version('scanpy')` instead
  if Version(scanpy.__version__) < Version("1.10"):
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:1598: FutureWarning: From 0.4 .update() will not pull obs/var columns from individual modalities by default anymore. Set mudata.set_options(pull_on_update=False) to adopt the new behaviour, which will become the default. Use new pull_obs/pull_var and push_obs/push_var methods for more flexibility.
  self._update_attr("var", axis=0, join_common=join_common)
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:963: UserWarning: Cannot join columns with the same name because var_names are intersecting.
  warnings.warn(
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:1461: FutureWarning: From 0.4 .update() will not pull obs/var columns from individual modalities by default anymore. Set mudata.set_options(pull_on_update=False) to adopt the new behaviour, which will become the default. Use new pull_obs/pull_var and push_obs/push_var methods for more flexibility.
  self._update_attr("obs", axis=1, join_common=join_common)
/home/sam/anaconda3/lib/python3.13/functools.py:934: ImplicitModificationWarning: Transforming to str index.
  return dispatch(args[0].__class__)(*args, **kw)
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:1598: FutureWarning: From 0.4 .update() will not pull obs/var columns from individual modalities by default anymore. Set mudata.set_options(pull_on_update=False) to adopt the new behaviour, which will become the default. Use new pull_obs/pull_var and push_obs/push_var methods for more flexibility.
  self._update_attr("var", axis=0, join_common=join_common)
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:963: UserWarning: Cannot join columns with the same name because var_names are intersecting.
  warnings.warn(
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:1461: FutureWarning: From 0.4 .update() will not pull obs/var columns from individual modalities by default anymore. Set mudata.set_options(pull_on_update=False) to adopt the new behaviour, which will become the default. Use new pull_obs/pull_var and push_obs/push_var methods for more flexibility.
  self._update_attr("obs", axis=1, join_common=join_common)
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:1598: FutureWarning: From 0.4 .update() will not pull obs/var columns from individual modalities by default anymore. Set mudata.set_options(pull_on_update=False) to adopt the new behaviour, which will become the default. Use new pull_obs/pull_var and push_obs/push_var methods for more flexibility.
  self._update_attr("var", axis=0, join_common=join_common)
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:963: UserWarning: Cannot join columns with the same name because var_names are intersecting.
  warnings.warn(
/home/sam/anaconda3/lib/python3.13/site-packages/mudata/_core/mudata.py:1461: FutureWarning: From 0.4 .update() will not pull obs/var columns from individual modalities by default anymore. Set mudata.set_options(pull_on_update=False) to adopt the new behaviour, which will become the default. Use new pull_obs/pull_var and push_obs/push_var methods for more flexibility.
  self._update_attr("obs", axis=1, join_common=join_common)
[1/4] Loading + sampling (50,000 cells)...
[2/4] Building Reactome map...
[3/4] Building feature matrix (FEATURE_SET='mix')...
  Feature matrix: 9661 features  (mix)
  adata_train: (50000, 9661)  pathway_map: (23859, 4): all the possible paths.
[4/4] Building graphs...
  Building protein-only reference graph...
  Protein features matched to Reactome: 164

=== Benchmark 1: BINN GCN (Reactome mask, mix) — 3 trials ===

[BINN donor CV — BINN GCN (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7009  val_loss=0.6881  patience=50
  epoch 40/400  train_loss=0.0053  val_loss=0.6951  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8000  val acc: 0.8571  test acc: 0.7778
    test acc=0.778  auc=0.857
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6891  val_loss=0.7115  patience=50
  epoch 40/400  train_loss=0.0034  val_loss=0.7003  patience=34
  epoch 80/400  train_loss=0.0001  val_loss=0.6748  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.6250  patience=48
  epoch 160/400  train_loss=0.0000  val_loss=0.5184  patience=50
  epoch 200/400  train_loss=0.0000  val_loss=0.5409  patience=14
  early stop at epoch 214
  donor-level fit acc: 1.0000  val acc: 0.5714  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6898  val_loss=0.6942  patience=50
  epoch 40/400  train_loss=0.0018  val_loss=0.6604  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.5794  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=1.0348  patience=11
  early stop at epoch 131
  donor-level fit acc: 1.0000  val acc: 0.5714  test acc: 1.0000
    test acc=1.000  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6798  val_loss=0.6912  patience=50
  epoch 40/400  train_loss=0.0039  val_loss=0.7122  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.417
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6944  val_loss=0.6872  patience=50
  epoch 40/400  train_loss=0.0025  val_loss=0.6906  patience=29
  epoch 80/400  train_loss=0.0001  val_loss=0.4716  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.4402  patience=32
  early stop at epoch 152
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=1.000
  [BINN GCN (mask, mix)]  acc=0.856±0.088  auc=0.826±0.214
  trial 1/3  acc=0.856  auc=0.826

[BINN donor CV — BINN GCN (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6804  val_loss=0.6901  patience=50
  epoch 40/400  train_loss=0.0012  val_loss=0.6823  patience=39
  epoch 80/400  train_loss=0.0002  val_loss=0.8518  patience=33
  early stop at epoch 113
  donor-level fit acc: 1.0000  val acc: 0.4286  test acc: 0.5556
    test acc=0.556  auc=0.571
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6885  val_loss=0.6840  patience=50
  epoch 40/400  train_loss=0.0027  val_loss=0.6831  patience=25
  epoch 80/400  train_loss=0.0000  val_loss=0.3843  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.1428  patience=34
  early stop at epoch 154
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6765  val_loss=0.7067  patience=50
  epoch 40/400  train_loss=0.0026  val_loss=0.8403  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.2308  val acc: 0.1429  test acc: 0.1250
    test acc=0.125  auc=0.571
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6930  val_loss=0.6867  patience=50
  epoch 40/400  train_loss=0.0034  val_loss=0.6845  patience=44
  epoch 80/400  train_loss=0.0001  val_loss=0.5902  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.5154  patience=34
  early stop at epoch 154
  donor-level fit acc: 1.0000  val acc: 0.5714  test acc: 0.8750
    test acc=0.875  auc=0.917
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6964  val_loss=0.6945  patience=50
  epoch 40/400  train_loss=0.0023  val_loss=0.6858  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.5519  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.6556  patience=25
  early stop at epoch 145
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.7500
    test acc=0.750  auc=1.000
  [BINN GCN (mask, mix)]  acc=0.636±0.281  auc=0.783±0.179
  trial 2/3  acc=0.636  auc=0.783

[BINN donor CV — BINN GCN (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6914  val_loss=0.6798  patience=50
  epoch 40/400  train_loss=0.0023  val_loss=0.6790  patience=34
  epoch 80/400  train_loss=0.0001  val_loss=0.6595  patience=43
  epoch 120/400  train_loss=0.0000  val_loss=1.0166  patience=3
  early stop at epoch 123
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.8889
    test acc=0.889  auc=0.857
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7095  val_loss=0.6815  patience=50
  epoch 40/400  train_loss=0.0028  val_loss=0.6654  patience=50
  epoch 80/400  train_loss=0.0000  val_loss=0.4855  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.0310  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.0188  patience=25
  epoch 200/400  train_loss=0.0000  val_loss=0.0311  patience=18
  early stop at epoch 218
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6962  val_loss=0.6871  patience=50
  epoch 40/400  train_loss=0.0011  val_loss=0.7248  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6972  val_loss=0.6820  patience=50
  epoch 40/400  train_loss=0.0031  val_loss=0.6796  patience=42
  epoch 80/400  train_loss=0.0001  val_loss=0.5681  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.1010  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.2023  patience=12
  early stop at epoch 172
  donor-level fit acc: 0.5385  val acc: 0.1429  test acc: 0.5000
    test acc=0.500  auc=1.000
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7019  val_loss=0.7004  patience=50
  epoch 40/400  train_loss=0.0020  val_loss=0.7366  patience=18
  early stop at epoch 58
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.667
  [BINN GCN (mask, mix)]  acc=0.778±0.148  auc=0.905±0.131
  trial 3/3  acc=0.778  auc=0.905

=== Benchmark 2: GCN unconstrained (no mask, mix) — 3 trials ===

[BINN donor CV — GCN (no mask, mix)]  donors=41  folds=5  pathway_mask=False
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6900  val_loss=0.6843  patience=50
  epoch 40/400  train_loss=0.0006  val_loss=0.4741  patience=48
  epoch 80/400  train_loss=0.0000  val_loss=0.2822  patience=41
  epoch 120/400  train_loss=0.0000  val_loss=0.2414  patience=34
  early stop at epoch 154
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8889
    test acc=0.889  auc=1.000
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6965  val_loss=0.6674  patience=50
  epoch 40/400  train_loss=0.0003  val_loss=0.7218  patience=27
  epoch 80/400  train_loss=0.0001  val_loss=0.2263  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.0428  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.0638  patience=37
  epoch 200/400  train_loss=0.0000  val_loss=0.0354  patience=21
  early stop at epoch 221
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.7500
    test acc=0.750  auc=0.857
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7006  val_loss=0.6742  patience=50
  epoch 40/400  train_loss=0.0004  val_loss=0.3204  patience=50
  epoch 80/400  train_loss=0.0000  val_loss=0.5804  patience=15
  early stop at epoch 95
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7042  val_loss=0.6767  patience=50
  epoch 40/400  train_loss=0.0005  val_loss=0.1522  patience=38
  epoch 80/400  train_loss=0.0000  val_loss=0.0035  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.0021  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.0014  patience=48
  epoch 200/400  train_loss=0.0000  val_loss=0.0011  patience=47
  epoch 240/400  train_loss=0.0000  val_loss=0.0013  patience=11
  early stop at epoch 251
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6979  val_loss=0.6821  patience=50
  epoch 40/400  train_loss=0.0005  val_loss=0.3118  patience=47
  epoch 80/400  train_loss=0.0001  val_loss=0.3280  patience=29
  epoch 120/400  train_loss=0.0000  val_loss=0.2881  patience=40
  epoch 160/400  train_loss=0.0000  val_loss=0.3673  patience=0
  early stop at epoch 160
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.917
  [GCN (no mask, mix)]  acc=0.828±0.064  auc=0.955±0.059
  trial 1/3  acc=0.828  auc=0.955

[BINN donor CV — GCN (no mask, mix)]  donors=41  folds=5  pathway_mask=False
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7123  val_loss=0.7010  patience=50
  epoch 40/400  train_loss=0.0006  val_loss=0.4160  patience=50
  epoch 80/400  train_loss=0.0000  val_loss=0.4563  patience=12
  early stop at epoch 92
  donor-level fit acc: 0.9600  val acc: 0.7143  test acc: 0.7778
    test acc=0.778  auc=0.857
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6863  val_loss=0.6721  patience=50
  epoch 40/400  train_loss=0.0008  val_loss=0.5999  patience=29
  epoch 80/400  train_loss=0.0001  val_loss=0.0089  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.0211  patience=24
  early stop at epoch 144
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6777  val_loss=0.6617  patience=50
  epoch 40/400  train_loss=0.0004  val_loss=0.6702  patience=38
  early stop at epoch 78
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 1.0000
    test acc=1.000  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6973  val_loss=0.6735  patience=50
  epoch 40/400  train_loss=0.0006  val_loss=0.5910  patience=30
  epoch 80/400  train_loss=0.0001  val_loss=0.1963  patience=47
  epoch 120/400  train_loss=0.0000  val_loss=0.3159  patience=7
  early stop at epoch 127
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.6250
    test acc=0.625  auc=0.833
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6730  val_loss=0.6850  patience=50
  epoch 40/400  train_loss=0.0005  val_loss=1.1995  patience=29
  early stop at epoch 69
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=1.000
  [GCN (no mask, mix)]  acc=0.806±0.126  auc=0.938±0.076
  trial 2/3  acc=0.806  auc=0.938

[BINN donor CV — GCN (no mask, mix)]  donors=41  folds=5  pathway_mask=False
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6893  val_loss=0.6822  patience=50
  epoch 40/400  train_loss=0.0006  val_loss=0.4875  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.4271  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.3505  patience=35
  early stop at epoch 155
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 1.0000
    test acc=1.000  auc=1.000
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7011  val_loss=0.6630  patience=50
  epoch 40/400  train_loss=0.0003  val_loss=0.4374  patience=50
  epoch 80/400  train_loss=0.0000  val_loss=0.0304  patience=44
  epoch 120/400  train_loss=0.0000  val_loss=0.0455  patience=4
  early stop at epoch 124
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7107  val_loss=0.6620  patience=50
  epoch 40/400  train_loss=0.0005  val_loss=1.2169  patience=32
  early stop at epoch 72
  donor-level fit acc: 0.9615  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6853  val_loss=0.6796  patience=50
  epoch 40/400  train_loss=0.0004  val_loss=0.5898  patience=27
  epoch 80/400  train_loss=0.0000  val_loss=0.2234  patience=36
  epoch 120/400  train_loss=0.0000  val_loss=0.2589  patience=38
  early stop at epoch 158
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6892  val_loss=0.6750  patience=50
  epoch 40/400  train_loss=0.0011  val_loss=0.5180  patience=37
  early stop at epoch 77
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.833
  [GCN (no mask, mix)]  acc=0.875±0.079  auc=0.938±0.076
  trial 3/3  acc=0.875  auc=0.938

=== Benchmark 3: Protein-only GCN — 3 trials ===

[BINN donor CV — GCN protein-only (Reactome mask)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7004  val_loss=0.6927  patience=50
  epoch 40/400  train_loss=0.2617  val_loss=0.6794  patience=50
  epoch 80/400  train_loss=0.0027  val_loss=0.6071  patience=49
  epoch 120/400  train_loss=0.0003  val_loss=2.0935  patience=14
  early stop at epoch 134
  donor-level fit acc: 0.9600  val acc: 0.5714  test acc: 0.8889
    test acc=0.889  auc=0.714
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6893  val_loss=0.7093  patience=50
  epoch 40/400  train_loss=0.2609  val_loss=0.7147  patience=25
  early stop at epoch 65
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.143
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7010  val_loss=0.6911  patience=50
  epoch 40/400  train_loss=0.3031  val_loss=0.7666  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8077  val acc: 0.5714  test acc: 0.6250
    test acc=0.625  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6691  val_loss=0.7007  patience=50
  epoch 40/400  train_loss=0.2492  val_loss=0.6842  patience=50
  epoch 80/400  train_loss=0.0078  val_loss=1.3951  patience=29
  early stop at epoch 109
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.750
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7041  val_loss=0.6850  patience=50
  epoch 40/400  train_loss=0.2466  val_loss=0.6925  patience=41
  epoch 80/400  train_loss=0.0020  val_loss=1.7191  patience=1
  early stop at epoch 81
  donor-level fit acc: 0.1923  val acc: 0.1429  test acc: 0.2500
    test acc=0.250  auc=0.917
  [GCN protein-only (Reactome mask)]  acc=0.678±0.234  auc=0.705±0.300
  trial 1/3  acc=0.678  auc=0.705

[BINN donor CV — GCN protein-only (Reactome mask)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6941  val_loss=0.6869  patience=50
  epoch 40/400  train_loss=0.1294  val_loss=0.7392  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8000  val acc: 0.8571  test acc: 0.7778
    test acc=0.778  auc=0.214
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6927  val_loss=0.6749  patience=50
  epoch 40/400  train_loss=0.2578  val_loss=0.6984  patience=11
  epoch 80/400  train_loss=0.0011  val_loss=0.1643  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.1855  patience=14
  early stop at epoch 134
  donor-level fit acc: 0.6538  val acc: 0.2857  test acc: 0.5000
    test acc=0.500  auc=1.000
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6806  val_loss=0.6945  patience=50
  epoch 40/400  train_loss=0.3138  val_loss=0.6745  patience=46
  epoch 80/400  train_loss=0.0057  val_loss=1.5751  patience=38
  early stop at epoch 118
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.500
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6923  val_loss=0.6912  patience=50
  epoch 40/400  train_loss=0.2736  val_loss=0.7216  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.750
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6859  val_loss=0.6840  patience=50
  epoch 40/400  train_loss=0.3098  val_loss=0.6740  patience=50
  epoch 80/400  train_loss=0.0337  val_loss=1.5129  patience=24
  early stop at epoch 104
  donor-level fit acc: 0.1923  val acc: 0.1429  test acc: 0.2500
    test acc=0.250  auc=0.500
  [GCN protein-only (Reactome mask)]  acc=0.631±0.227  auc=0.593±0.265
  trial 2/3  acc=0.631  auc=0.593

[BINN donor CV — GCN protein-only (Reactome mask)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6935  val_loss=0.6885  patience=50
  epoch 40/400  train_loss=0.2151  val_loss=0.6771  patience=44
  epoch 80/400  train_loss=0.0010  val_loss=2.0537  patience=27
  early stop at epoch 107
  donor-level fit acc: 0.2000  val acc: 0.1429  test acc: 0.2222
    test acc=0.222  auc=0.500
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7110  val_loss=0.7256  patience=50
  epoch 40/400  train_loss=0.3870  val_loss=0.6561  patience=50
  epoch 80/400  train_loss=0.0036  val_loss=0.8406  patience=18
  epoch 120/400  train_loss=0.0005  val_loss=1.8375  patience=23
  early stop at epoch 143
  donor-level fit acc: 0.5385  val acc: 0.2857  test acc: 0.5000
    test acc=0.500  auc=1.000
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6995  val_loss=0.6905  patience=50
  epoch 40/400  train_loss=0.2351  val_loss=0.8020  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7053  val_loss=0.7020  patience=50
  epoch 40/400  train_loss=0.2815  val_loss=0.6772  patience=50
  epoch 80/400  train_loss=0.0010  val_loss=1.2370  patience=10
  early stop at epoch 90
  donor-level fit acc: 0.1923  val acc: 0.1429  test acc: 0.2500
    test acc=0.250  auc=0.500
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7032  val_loss=0.6798  patience=50
  epoch 40/400  train_loss=0.2641  val_loss=0.7224  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=1.000
  [GCN protein-only (Reactome mask)]  acc=0.519±0.261  auc=0.600±0.374
  trial 3/3  acc=0.519  auc=0.600

=== Benchmark 4: BINN GAT (Reactome mask, mix) — 3 trials ===

[BINN donor CV — BINN GAT (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6906  val_loss=0.6802  patience=50
  epoch 40/400  train_loss=0.0024  val_loss=0.6899  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8000  val acc: 0.8571  test acc: 0.7778
    test acc=0.778  auc=0.857
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7121  val_loss=0.7070  patience=50
  epoch 40/400  train_loss=0.0033  val_loss=0.7172  patience=29
  epoch 80/400  train_loss=0.0002  val_loss=0.2827  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.4085  patience=16
  early stop at epoch 136
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7105  val_loss=0.6958  patience=50
  epoch 40/400  train_loss=0.0030  val_loss=0.7061  patience=25
  early stop at epoch 65
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7054  val_loss=0.6879  patience=50
  epoch 40/400  train_loss=0.0028  val_loss=0.6817  patience=47
  epoch 80/400  train_loss=0.0001  val_loss=0.2839  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.0080  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.0051  patience=49
  epoch 200/400  train_loss=0.0000  val_loss=0.0038  patience=41
  epoch 240/400  train_loss=0.0000  val_loss=0.0037  patience=46
  epoch 280/400  train_loss=0.0000  val_loss=0.0035  patience=30
  epoch 320/400  train_loss=0.0000  val_loss=0.0034  patience=30
  epoch 360/400  train_loss=0.0000  val_loss=0.0038  patience=29
  early stop at epoch 389
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.7500
    test acc=0.750  auc=0.750
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6792  val_loss=0.6920  patience=50
  epoch 40/400  train_loss=0.0038  val_loss=0.7042  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8077  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.333
  [BINN GAT (mask, mix)]  acc=0.806±0.058  auc=0.731±0.203
  trial 1/3  acc=0.806  auc=0.731

[BINN donor CV — BINN GAT (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7006  val_loss=0.6911  patience=50
  epoch 40/400  train_loss=0.0042  val_loss=0.7707  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.8000  val acc: 0.8571  test acc: 0.7778
    test acc=0.778  auc=0.714
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6820  val_loss=0.6888  patience=50
  epoch 40/400  train_loss=0.0016  val_loss=0.7307  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.000
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6846  val_loss=0.6991  patience=50
  epoch 40/400  train_loss=0.0024  val_loss=0.7037  patience=27
  epoch 80/400  train_loss=0.0001  val_loss=0.6949  patience=39
  early stop at epoch 119
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.6250
    test acc=0.625  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7132  val_loss=0.6861  patience=50
  epoch 40/400  train_loss=0.0025  val_loss=0.6922  patience=11
  epoch 80/400  train_loss=0.0002  val_loss=0.4737  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.5358  patience=30
  early stop at epoch 150
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.833
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6842  val_loss=0.6904  patience=50
  epoch 40/400  train_loss=0.0021  val_loss=0.6817  patience=36
  epoch 80/400  train_loss=0.0002  val_loss=0.4961  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.3774  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.3273  patience=40
  epoch 200/400  train_loss=0.0000  val_loss=0.3233  patience=15
  early stop at epoch 215
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=1.000
  [BINN GAT (mask, mix)]  acc=0.781±0.093  auc=0.710±0.371
  trial 2/3  acc=0.781  auc=0.710

[BINN donor CV — BINN GAT (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7019  val_loss=0.6867  patience=50
  epoch 40/400  train_loss=0.0033  val_loss=0.6833  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.6758  patience=49
  epoch 120/400  train_loss=0.0001  val_loss=0.9748  patience=9
  early stop at epoch 129
  donor-level fit acc: 1.0000  val acc: 0.5714  test acc: 1.0000
    test acc=1.000  auc=1.000
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7078  val_loss=0.7264  patience=50
  epoch 40/400  train_loss=0.0037  val_loss=0.7357  patience=22
  early stop at epoch 62
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7008  val_loss=0.6849  patience=50
  epoch 40/400  train_loss=0.0026  val_loss=0.7021  patience=21
  early stop at epoch 61
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7041  val_loss=0.6890  patience=50
  epoch 40/400  train_loss=0.0039  val_loss=0.6782  patience=49
  epoch 80/400  train_loss=0.0001  val_loss=0.5234  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.1237  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.1015  patience=32
  epoch 200/400  train_loss=0.0000  val_loss=0.0958  patience=28
  epoch 240/400  train_loss=0.0000  val_loss=0.0681  patience=50
  epoch 280/400  train_loss=0.0000  val_loss=0.0724  patience=10
  early stop at epoch 290
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6944  val_loss=0.7070  patience=50
  epoch 40/400  train_loss=0.0032  val_loss=0.7373  patience=20
  early stop at epoch 60
  donor-level fit acc: 0.8846  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.667
  [BINN GAT (mask, mix)]  acc=0.875±0.079  auc=0.905±0.131
  trial 3/3  acc=0.875  auc=0.905

=== Benchmark 5: BINN ANN (Reactome mask, mix) — 3 trials ===

[BINN donor CV — BINN ANN (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6988  val_loss=0.6842  patience=50
  epoch 40/400  train_loss=0.0019  val_loss=0.6741  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.4739  patience=50
  epoch 120/400  train_loss=0.0001  val_loss=0.3694  patience=50
  epoch 160/400  train_loss=0.0001  val_loss=0.3668  patience=21
  early stop at epoch 181
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8889
    test acc=0.889  auc=1.000
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6824  val_loss=0.7029  patience=50
  epoch 40/400  train_loss=0.0045  val_loss=0.7323  patience=15
  early stop at epoch 55
  donor-level fit acc: 0.2308  val acc: 0.1429  test acc: 0.1250
    test acc=0.125  auc=0.857
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6975  val_loss=0.6907  patience=50
  epoch 40/400  train_loss=0.0031  val_loss=0.6597  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.6083  patience=44
  epoch 120/400  train_loss=0.0001  val_loss=0.6860  patience=4
  early stop at epoch 124
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.857
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6743  val_loss=0.6931  patience=50
  epoch 40/400  train_loss=0.0015  val_loss=0.6987  patience=24
  epoch 80/400  train_loss=0.0002  val_loss=0.5381  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.4478  patience=43
  epoch 160/400  train_loss=0.0000  val_loss=0.4231  patience=49
  epoch 200/400  train_loss=0.0000  val_loss=0.4266  patience=9
  epoch 240/400  train_loss=0.0000  val_loss=0.4283  patience=27
  early stop at epoch 267
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.6250
    test acc=0.625  auc=0.833
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7002  val_loss=0.6911  patience=50
  epoch 40/400  train_loss=0.0012  val_loss=0.6832  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.5962  patience=47
  epoch 120/400  train_loss=0.0001  val_loss=0.6359  patience=7
  early stop at epoch 127
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.833
  [BINN ANN (mask, mix)]  acc=0.653±0.281  auc=0.876±0.063
  trial 1/3  acc=0.653  auc=0.876

[BINN donor CV — BINN ANN (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6965  val_loss=0.6890  patience=50
  epoch 40/400  train_loss=0.0023  val_loss=0.6775  patience=48
  epoch 80/400  train_loss=0.0001  val_loss=0.7377  patience=35
  early stop at epoch 115
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7778
    test acc=0.778  auc=0.714
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7034  val_loss=0.7057  patience=50
  epoch 40/400  train_loss=0.0027  val_loss=0.6849  patience=36
  epoch 80/400  train_loss=0.0001  val_loss=0.4923  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.3880  patience=50
  epoch 160/400  train_loss=0.0000  val_loss=0.3723  patience=39
  epoch 200/400  train_loss=0.0000  val_loss=0.3900  patience=32
  epoch 240/400  train_loss=0.0000  val_loss=0.3871  patience=39
  epoch 280/400  train_loss=0.0000  val_loss=0.3564  patience=28
  epoch 320/400  train_loss=0.0000  val_loss=0.3602  patience=17
  early stop at epoch 337
  donor-level fit acc: 1.0000  val acc: 1.0000  test acc: 0.6250
    test acc=0.625  auc=0.714
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6990  val_loss=0.7048  patience=50
  epoch 40/400  train_loss=0.0018  val_loss=0.6922  patience=31
  epoch 80/400  train_loss=0.0003  val_loss=0.5053  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.5116  patience=10
  early stop at epoch 130
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=1.000
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.7180  val_loss=0.6823  patience=50
  epoch 40/400  train_loss=0.0034  val_loss=0.6840  patience=29
  epoch 80/400  train_loss=0.0001  val_loss=0.6230  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.6854  patience=16
  early stop at epoch 136
  donor-level fit acc: 1.0000  val acc: 0.7143  test acc: 0.7500
    test acc=0.750  auc=1.000
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6915  val_loss=0.6937  patience=50
  epoch 40/400  train_loss=0.0011  val_loss=0.6731  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.6042  patience=43
  epoch 120/400  train_loss=0.0000  val_loss=1.4945  patience=3
  early stop at epoch 123
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=1.000
  [BINN ANN (mask, mix)]  acc=0.756±0.080  auc=0.886±0.140
  trial 2/3  acc=0.756  auc=0.886

[BINN donor CV — BINN ANN (mask, mix)]  donors=41  folds=5  pathway_mask=True
  fold 1/5  train=32  test=9

[Donor attention ensemble] donors=41, fit=25, val=7, test=9, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6938  val_loss=0.6830  patience=50
  epoch 40/400  train_loss=0.0021  val_loss=0.6738  patience=39
  epoch 80/400  train_loss=0.0002  val_loss=0.6266  patience=42
  epoch 120/400  train_loss=0.0000  val_loss=0.9059  patience=2
  early stop at epoch 122
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7778
    test acc=0.778  auc=0.929
  fold 2/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6946  val_loss=0.6719  patience=50
  epoch 40/400  train_loss=0.0016  val_loss=0.6767  patience=11
  early stop at epoch 51
  donor-level fit acc: 0.7692  val acc: 0.8571  test acc: 0.8750
    test acc=0.875  auc=0.143
  fold 3/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6886  val_loss=0.6822  patience=50
  epoch 40/400  train_loss=0.0025  val_loss=0.6793  patience=31
  epoch 80/400  train_loss=0.0001  val_loss=0.5311  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.7037  patience=12
  early stop at epoch 132
  donor-level fit acc: 1.0000  val acc: 0.5714  test acc: 0.5000
    test acc=0.500  auc=0.429
  fold 4/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6982  val_loss=0.6901  patience=50
  epoch 40/400  train_loss=0.0019  val_loss=0.6828  patience=50
  epoch 80/400  train_loss=0.0001  val_loss=0.7137  patience=38
  early stop at epoch 118
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.917
  fold 5/5  train=33  test=8

[Donor attention ensemble] donors=41, fit=26, val=7, test=8, embed_dim=32, fusion_in=99, fusion_dropout=0.3, class_weights=True, early_stop=True, classes=2, epochs=400
  epoch 1/400  train_loss=0.6991  val_loss=0.6926  patience=50
  epoch 40/400  train_loss=0.0030  val_loss=0.6942  patience=26
  epoch 80/400  train_loss=0.0001  val_loss=0.6640  patience=50
  epoch 120/400  train_loss=0.0000  val_loss=0.5399  patience=34
  early stop at epoch 154
  donor-level fit acc: 1.0000  val acc: 0.8571  test acc: 0.7500
    test acc=0.750  auc=0.917
  [BINN ANN (mask, mix)]  acc=0.731±0.124  auc=0.667±0.324
  trial 3/3  acc=0.731  auc=0.667

========================================================================
TIER-2 SUMMARY  (50,000 cells, 5-fold donor CV, feature_set='mix')
========================================================================
  Model                                                 Acc         AUC
  ---------------------------------------------- ----------  ----------
  BINN GCN (Reactome mask, mix)                  0.756±0.091  0.838±0.050
  GCN unconstrained (no mask, mix)               0.836±0.029  0.944±0.008
  GCN protein-only (Reactome proteins)           0.609±0.066  0.633±0.051
  BINN GAT (Reactome mask, mix)                  0.820±0.040  0.782±0.087
  BINN ANN (Reactome mask, mix)                  0.713±0.044  0.810±0.101
========================================================================