Experiment: polyatomic+polyatomic on freesolv

FOLD 1/5
GNN Train Loss: 0.1480
RMSE: 2.4752, MAE: 2.0829, R2: 0.5170

FOLD 2/5
GNN Train Loss: 0.1089
RMSE: 1.5965, MAE: 1.0820, R2: 0.7859

FOLD 3/5
GNN Train Loss: 0.1446
RMSE: 1.7065, MAE: 1.2664, R2: 0.7602

FOLD 4/5
GNN Train Loss: 0.1632
RMSE: 2.0845, MAE: 1.4736, R2: 0.6645

FOLD 5/5
GNN Train Loss: 0.1325
RMSE: 1.8133, MAE: 1.2575, R2: 0.7322

====== K-FOLD SUMMARY ======
RMSE: 1.9365  [1.6839, 2.2646]
MAE: 1.4233  [1.1891, 1.7618]
R2: 0.6924  [0.6030, 0.7649]

★ Using fold 2 model for test inference
RMSE: 1.6293, MAE: 1.1042, R2: 0.7859

====== HELD-OUT TEST METRICS ======
{'rmse': 1.6293278795693167, 'mae': 1.104212760925293, 'r2': 0.7858816385269165}
Test RMSE: 1.6139  (95 % CI: 1.2757–1.9908)

Test MAE : 1.1024  (95 % CI: 0.8973–1.3226)

Test R²  : 0.7810  (95 % CI: 0.6604–0.8753)