Titer prediction on Literature-mined fermentation records n=410 unique experiments (test)
0.413R2CANOPY (3B)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| CANOPY (3B)Model scale=3B, Backbone=heterogeneous HGT, Augmentations=SignNet/JK/virtual-node2026.07 | 0.413 | 82 | |
| CANOPY (500M)Model scale=500M, Backbone=heterogeneous HGT, Augmentations=SignNet/JK/virtual-node2026.07 | 0.38 | 80.5 | |
| GraphSAGE + SN/JK/VNBackbone=homogeneous SAGEConv, Augmentations=SignNet/JK/virtual-node2026.07 | 0.334 | 73.4 | |
| HGTBackbone=heterogeneous HGT, Augmentations=vanilla2026.07 | 0.308 | 71.1 | |
| CANOPY (Demo)Model scale=Demo, Backbone=heterogeneous HGT, Augmentations=SignNet/JK/virtual-node2026.07 | 0.301 | 70.3 | |
| GraphSAGEBackbone=homogeneous SAGEConv, Augmentations=vanilla2026.07 | 0.241 | 71.3 | |
| XGBoostInput features=raw conditions, Feature dimensionality=429-dim, Configuration=500 trees, depth 62026.07 | 0.236 | 72.6 | |
| RidgeInput features=raw conditions, Feature dimensionality=429-dim2026.07 | 0.086 | 69.8 | |
| MLPInput features=raw conditions, Architecture=two hidden layers, 128→64, Feature dimensionality=429-dim2026.07 | 0.078 | 69.1 |