Table 1 Protein family-based average Spearman scores of the best models and baseline models in each dataset split
From: How to approach machine learning-based prediction of drug/compound–target interactions
Name of the descriptor set/representation (explanation) | Fully-dissimilar-split | Dissimilar-compound-split | Random- split |
|---|---|---|---|
Best performing protein representation (compound: ECFP4) | 0.363 | 0.518 | 0.868 |
random200 (protein: random continuous vectors, compound: ECFP4) | 0.193 | 0.436 | 0.861 |
only-ecfp4 (no protein vector, compound: ECFP4) | 0.302 | 0.379 | 0.709 |
random200-random-ecfp4 (protein: random continuous vectors, compound: random binary vectors) | 0.056 | 0.272 | 0.504 |
only-random-ecfp4 (no protein vector, compound: random binary vectors) | 0.002 | − 0.002 | 0.315 |