Method: BLOSUM62 Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.14 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source BLOSUM62 on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(BLOSUM62), column(2-vs-rest) Method: CNN Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.32 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source CNN on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(CNN), column(2-vs-rest) Configuration: ESM-1b (mean) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.36 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-1b (mean) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-1b (mean)), column(2-vs-rest) Configuration: ESM-1b (mut mean) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.19 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-1b (mut mean) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-1b (mut mean)), column(2-vs-rest) Configuration: ESM-1b (per AA) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.55 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-1b (per AA) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-1b (per AA)), column(2-vs-rest) Configuration: ESM-1v (mean) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.32 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-1v (mean) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-1v (mean)), column(2-vs-rest) Configuration: ESM-1v (mut mean) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.19 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-1v (mut mean) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-1v (mut mean)), column(2-vs-rest) Configuration: ESM-1v (per AA) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.28 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-1v (per AA) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-1v (per AA)), column(2-vs-rest) Configuration: ESM-untrained (mean) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.05 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-untrained (mean) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-untrained (mean)), column(2-vs-rest) Configuration: ESM-untrained (mut mean) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.21 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-untrained (mut mean) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-untrained (mut mean)), column(2-vs-rest) Configuration: ESM-untrained (per AA) Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.06 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source ESM-untrained (per AA) on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(ESM-untrained (per AA)), column(2-vs-rest) Method: Levenshtein Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.16 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source Levenshtein on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(Levenshtein), column(2-vs-rest) Method: Ridge Task: FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.59 Spearmanunitless · higher
Uncertainty: Not yet extracted
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Methods, coverage and source Ridge on FLIP GB1-2-VS-REST: GB1 fitness prediction, 2-vs-rest split
Trained on the split's training set with ten percent held out for validation, scored on its test set.
Aggregation: Not reported
flip primary benchmark evidence · Table 4, row(Ridge), column(2-vs-rest) Configuration: AMPLIFY-350M (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.497 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'AMPLIFY-350M' (page 12 of 13) Configuration: Ankh-base (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.617 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 Ankh-base on gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'GB1 two-vs-rest', column 'Ankh-base' (page 13 of 13) Configuration: Ankh3-large (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.299 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 Ankh3-large on gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'GB1 two-vs-rest', column 'Ankh3-large' (page 13 of 13) Configuration: DPLM-650M (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.584 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'DPLM-650M' (page 12 of 13) Configuration: E1-600M (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.66 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'E1-600M' (page 12 of 13) Configuration: ESM-1b (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.615 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'ESM-1b' (page 12 of 13) Configuration: ESM-2 650M (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.552 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'ESM-2 650M' (page 12 of 13) Configuration: ESM C 600M (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.64 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'ESM C 600M' (page 12 of 13) Configuration: ESM3-open (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.519 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'GB1 two-vs-rest', column 'ESM3-open' (page 13 of 13) Configuration: HyPLM-Flash, BF16 evaluation (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.628 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 6, row 'GB1 two-vs-rest', column 'HyPLM-Flash' (page 12 of 13) Configuration: ProtBERT (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.507 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'GB1 two-vs-rest', column 'ProtBERT' (page 13 of 13) Configuration: ProtT5-XL (HyPLM report) Protocol: GB1 two-vs-rest (HyPLM report) Dataset subset: FLIP GB1, 2-vs-rest split (FLIP split) 0.414 Spearman (Spearman correlation on the FLIP GB1 two-vs-rest split)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 ProtT5-XL on gb1-two-vs-rest (HyPLM report)
hyplm-flash-20261106-protocol-gb1-two-vs-rest
Aggregation: Not reported
HyPLM Technical Report (Hyra Team, 2026-10-12) · Table 7, row 'GB1 two-vs-rest', column 'ProtT5-XL' (page 13 of 13)