We trained a 1.3B parameter version of SaProt. Results showed on par performance between SaProt 1.3B and 650M, suggesting that increasing model size alone may not significantly improve performance. We welcome more evaluations by the community.
| model | ClinVar | ProteinGym | Mega-scale |
|---|---|---|---|
| AUC | Spearman's ρ | Spearman's ρ | |
| SaProt (650M) | 0.909 | 0.457 | 0.574 |
| SaProt (1.3B) | 0.910 | 0.460 | 0.588 |
- Downloads last month
- 12