Experiment: polyatomic+polyatomic on lipophil

FOLD 1/5
GNN Train Loss: 0.4286
RMSE: 0.8033, MAE: 0.6121, R2: 0.5614

FOLD 2/5
GNN Train Loss: 0.4082
RMSE: 0.8277, MAE: 0.6283, R2: 0.5242

FOLD 3/5
GNN Train Loss: 0.4646
RMSE: 1.0686, MAE: 0.8525, R2: 0.2336

FOLD 4/5
GNN Train Loss: 0.4228
RMSE: 0.8646, MAE: 0.6720, R2: 0.4946

FOLD 5/5
GNN Train Loss: 0.4391
RMSE: 0.8195, MAE: 0.6323, R2: 0.5503

====== K-FOLD SUMMARY ======
RMSE: 0.8743  [0.8147, 0.9748]
MAE: 0.6832  [0.6226, 0.7724]
R2: 0.4736  [0.3514, 0.5495]

★ Using fold 1 model for test inference
RMSE: 0.8043, MAE: 0.6128, R2: 0.5614

====== HELD-OUT TEST METRICS ======
{'rmse': 0.804273702649856, 'mae': 0.6127852201461792, 'r2': 0.5613847374916077}
Test RMSE: 0.8041  (95 % CI: 0.7516–0.8585)

Test MAE : 0.6136  (95 % CI: 0.5733–0.6522)

Test R²  : 0.5617  (95 % CI: 0.4998–0.6203)