Split novelty, full
Held-out molecule novelty¶
Maximum Morgan(r=2) Tanimoto to any of 2,618,349 sampled training molecules, over 400 held-out molecules.
| statistic | value |
|---|---|
| median | 0.806 |
| 25th percentile | 0.729 |
| 75th percentile | 0.868 |
| 95th percentile | 1.000 |
| share with a near-twin (>= 0.9) | 0.158 |
| share with a close analogue (>= 0.7) | 0.833 |
| share genuinely novel (< 0.5) | 0.007 |
A high near-twin share would mean exact-match scores reflect interpolation between close analogues rather than recognition of novel structures.