Gene deletion fitness prediction on Yeast Digenic Deletion Fitness (10-fold cross-validation)
0.377Mean R2GNN + Ldistance
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GNN + LdistanceModel Architecture=GNN, Embedding Type=Pre-trained box embeddings, Loss Function=Distance-based semantic loss2026.05 | 0.377 | 0.046 | |
| GNN + LoverlapModel Architecture=GNN, Embedding Type=Pre-trained box embeddings, Loss Function=Overlap semantic loss2026.05 | 0.368 | 0.038 | |
| GNN with prior box embeddingsModel Architecture=GNN, Embedding Type=Pre-trained box embeddings2026.05 | 0.36 | 0.043 | |
| GNN with subClassOf-links in KGModel Architecture=GNN, Embedding Type=Task-specific shallow embeddings, Hierarchy=Introduced as links in KG2026.05 | 0.35 | 0.049 | |
| GNN without box embeddingsModel Architecture=GNN, Embedding Type=Task-specific shallow embeddings (non-box form)2026.05 | 0.348 | 0.05 | |
| Instantiations + LightGBMModel Architecture=LightGBM, Embedding Type=Instantiations, KG Representation=Sparse feature matrix2026.05 | 0.211 | 0.022 | |
| ComplEx + LightGBMModel Architecture=LightGBM, Embedding Type=ComplEx, KG Representation=Sparse feature matrix2026.05 | 0.191 | 0.039 |