Configuration: AMPLIFY-350M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.363 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source AMPLIFY-350M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'AMPLIFY-350M' (page 12 of 13) Configuration: DPLM-3B (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.423 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source DPLM-3B on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 5, row 'DPLM-3B†', column 'Average' (page 11 of 13) Configuration: DPLM-650M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.41 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source DPLM-650M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'DPLM-650M' (page 12 of 13) Configuration: E1-150M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.401 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source E1-150M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 5, row 'E1-150M†', column 'Average' (page 11 of 13) Configuration: E1-300M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.417 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source E1-300M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 5, row 'E1-300M†', column 'Average' (page 11 of 13) Configuration: E1-600M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.42 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source E1-600M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'E1-600M' (page 12 of 13) Configuration: ESM-1b (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.389 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM-1b on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'ESM-1b' (page 12 of 13) Configuration: ESM-2 150M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.388 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM-2 150M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Figure 3a, bar 'ESM-2 150M', printed value label (page 5 of 13) Configuration: ESM-2 15B (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.403 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM-2 15B on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Figure 3a, bar 'ESM-2 15B', printed value label (page 5 of 13) Configuration: ESM-2 3B (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.407 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM-2 3B on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Figure 3a, bar 'ESM-2 3B', printed value label (page 5 of 13) Configuration: ESM-2 650M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.415 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM-2 650M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'ESM-2 650M' (page 12 of 13) Configuration: ESM C 300M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.406 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM C 300M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Figure 3a, bar 'ESM C 300M', printed value label (page 5 of 13) Configuration: ESM C 600M (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.406 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM C 600M on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'ESM C 600M' (page 12 of 13) Configuration: ESM C 6B (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.409 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM C 6B on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 5, row 'ESM C 6B†', column 'Average' (page 11 of 13) Configuration: ESM3-open (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.409 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ESM3-open on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'ProteinGym zero-shot', column 'ESM3-open' (page 13 of 13) Configuration: HyPLM-Flash, BF16 evaluation (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.409 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source HyPLM-Flash on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'ProteinGym zero-shot', column 'HyPLM-Flash' (page 12 of 13) Configuration: ProtBERT (HyPLM report) Protocol: ProteinGym zero-shot (HyPLM report) Dataset subset: ProteinGym DMS substitution assays at commit 144fe22 (217 assays) 0.343 Spearman (Spearman correlation averaged over the 217 substitution assays of ProteinGym v1.3 with the benchmark's official aggregation, from masked-marginal zero-shot scores; per-assay values rounded to three decimals before aggregation)unitless · higher
Uncertainty: Not reported by the source: Section 4 says intervals are paired bootstrap intervals; none is printed here.
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons The recent-PDB likelihood panel is described two ways. Section 4.3 and the Table 3 caption give 887 PDB chains released after 2 September 2026; Table 4 gives 887 RCSB PDB entries released 9 September to 7 October 2026. The start of the window and whether the unit is a chain or an entry both differ, so the panel cannot be rebuilt from the report.HyPLM Technical Report (Hyra Team, 2026-10-12) · Section 4.3; Table 3 caption; Table 4 row 'PDB time split'
Appendix A states that the authors could not reproduce xTrimoPGLM's published ProteinGym scores (their runs fall 0.07 to 0.09 below the leaderboard) and that its likelihood is implausibly high, pointing to an incompatibility between their adapter and its masked-language-model head. They withhold its zero-shot and likelihood results and keep its embedding-based results as unaffected, but those results come from the same adapter and that claim was not tested.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Deviations'
Appendix A states that for GB1 two-vs-rest, GB1 three-vs-rest and the cross-protein hydrolase split of FLIP2, validation data come from the training distribution while the test sets are strongly shifted, so probe selection is dominated by noise and numerically equivalent runs of the same model differ by up to 0.2. The authors print these cells but leave them out of every summary statistic.HyPLM Technical Report (Hyra Team, 2026-10-12) · Appendix A 'Statistics'
Methods, coverage and source ProtBERT on proteingym-zero-shot (HyPLM report)
hyplm-flash-20261106-protocol-proteingym-zero-shot
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'ProteinGym zero-shot', column 'ProtBERT' (page 13 of 13)