Where one study disagrees with what other studies report for the same models, and what we are doing to resolve each case.
These are automatic leads, not findings of error. Each release, a check compares how every study ranks the models it shares with other studies. It flags a study that reverses the order the others agree on, or whose gap to a shared reference model lies well outside theirs. Different tasks, populations or metrics often explain a flag.
A flag never hides a result. Flagged results stay in every table, page and download, marked Contested. A result is withheld only when a reviewer marks it disputed.
Brandes et al. 2023 Supplementary Table 2: full benchmark evaluations of ESM1b and all other methods ranks ESM-1b above EVE in 6 of 6 comparisons; 2 of 2 other studies rank EVE above ESM-1b.
2 papers rank EVE above ESM-1b in 22 of their 26 comparisons and ESM-1b higher in 1; 3 are ties. Brandes et al. 2023 ranks ESM-1b above EVE in all 6 comparisons that are not ties (2 are ties).
ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design ranks ESM-IF1 above ProteinMPNN in 19 of 19 comparisons; proteinbench primary benchmark evidence ranks ProteinMPNN above ESM-IF1 in 10 of 10 comparisons. No other study gives a verdict, so neither is the outlier.
2 papers rank ESM-IF1 above ProteinMPNN in 19 of their 29 comparisons and ProteinMPNN higher in 10. Of those papers, proteinbench primary benchmark evidence ranks ProteinMPNN higher on its own (10 of 10).
ProteinGym: Large-Scale Benchmarks for Protein Fitness Prediction and Design ranks ESM3 above ProteinMPNN in 18 of 19 comparisons; proteinbench primary benchmark evidence ranks ProteinMPNN above ESM3 in 10 of 10 comparisons. No other study gives a verdict, so neither is the outlier.
2 papers rank ESM3 above ProteinMPNN in 18 of their 29 comparisons and ProteinMPNN higher in 11. Of those papers, proteinbench primary benchmark evidence ranks ProteinMPNN higher on its own (10 of 10).