Experiment: polyatomic+polyatomic on bindingdb

FOLD 1/5
GNN Train Loss: 0.8965
RMSE: 1.5462, MAE: 1.2688, R2: 0.0660

FOLD 2/5
GNN Train Loss: 0.9111
RMSE: 1.5596, MAE: 1.2989, R2: 0.0780

FOLD 3/5
GNN Train Loss: 0.8896
RMSE: 1.6106, MAE: 1.3290, R2: 0.0211

FOLD 4/5
GNN Train Loss: 0.9180
RMSE: 1.6593, MAE: 1.3322, R2: -0.0659

FOLD 5/5
GNN Train Loss: 0.9116
RMSE: 1.5981, MAE: 1.3265, R2: 0.0310

====== K-FOLD SUMMARY ======
RMSE: 1.5943  [1.5619, 1.6296]
MAE: 1.3110  [1.2869, 1.3298]
R2: 0.0261  [-0.0201, 0.0638]

★ Using fold 1 model for test inference
RMSE: 1.5625, MAE: 1.2823, R2: 0.0660

====== HELD-OUT TEST METRICS ======
{'rmse': 1.5625251767991144, 'mae': 1.2822535037994385, 'r2': 0.0660201907157898}
Test RMSE: 1.5608  (95 % CI: 1.4350–1.6923)

Test MAE : 1.2846  (95 % CI: 1.1873–1.3814)

Test R²  : 0.0614  (95 % CI: -0.0021–0.1309)