Table 6. Model Ablationsa.
Ablation | Pearson’s r |
---|---|
Base model (FS-CAP) | 0.32 ± 0.01 |
No query encoding | 0.30 ± 0.01 |
Concatenated context | 0.28 ± 0.00 |
No context | 0.17 ± 0.03 |
Mean correlation between ground-truth and predicted activities across all test targets in BindingDB using 8 context compounds. We report the mean ± one standard deviation from three independent training runs with random seeds.