Configuration: CMLM (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.2 test (1,120 chains) 0.806 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source CMLM on CATH4.2 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'CMLM', column 'CATH4.2 test' (page 16 of 21) Configuration: CMLM (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.3 test (1,882 chains) 0.809 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source CMLM on CATH4.3 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'CMLM', column 'CATH4.3 test' (page 16 of 21) Configuration: CMLM (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: RecentPDB (121 chains from 111 recent PDB entries) 0.774 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source CMLM on RecentPDB (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'CMLM', column 'RecentPDB' (page 16 of 21) Configuration: CMLM (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS50 (50 chains) 0.871 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source CMLM on TS50 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'CMLM', column 'TS50' (page 16 of 21) Configuration: CMLM (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS500 (500 chains) 0.879 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source CMLM on TS500 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'CMLM', column 'TS500' (page 16 of 21) Configuration: DPLM-2 3B (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.2 test (1,120 chains) 0.809 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source DPLM-2 3B on CATH4.2 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'DPLM-2 3B', column 'CATH4.2 test' (page 16 of 21) Configuration: DPLM-2 3B (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.3 test (1,882 chains) 0.795 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source DPLM-2 3B on CATH4.3 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'DPLM-2 3B', column 'CATH4.3 test' (page 16 of 21) Configuration: DPLM-2 3B (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: RecentPDB (121 chains from 111 recent PDB entries) 0.765 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source DPLM-2 3B on RecentPDB (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'DPLM-2 3B', column 'RecentPDB' (page 16 of 21) Configuration: DPLM-2 3B (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS50 (50 chains) 0.87 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source DPLM-2 3B on TS50 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'DPLM-2 3B', column 'TS50' (page 16 of 21) Configuration: DPLM-2 3B (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS500 (500 chains) 0.869 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source DPLM-2 3B on TS500 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'DPLM-2 3B', column 'TS500' (page 16 of 21) Configuration: HyInv default (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.2 test (1,120 chains) 0.855 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source HyInv on CATH4.2 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'HyInv', column 'CATH4.2 test' (page 16 of 21) Configuration: HyInv default (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.3 test (1,882 chains) 0.84 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source HyInv on CATH4.3 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'HyInv', column 'CATH4.3 test' (page 16 of 21) Configuration: HyInv default (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: RecentPDB (121 chains from 111 recent PDB entries) 0.861 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source HyInv on RecentPDB (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'HyInv', column 'RecentPDB' (page 16 of 21) Configuration: HyInv default (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS50 (50 chains) 0.92 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source HyInv on TS50 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'HyInv', column 'TS50' (page 16 of 21) Configuration: HyInv default (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS500 (500 chains) 0.899 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source HyInv on TS500 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'HyInv', column 'TS500' (page 16 of 21) Configuration: LigandMPNN (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.2 test (1,120 chains) 0.841 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LigandMPNN on CATH4.2 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LigandMPNN', column 'CATH4.2 test' (page 16 of 21) Configuration: LigandMPNN (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.3 test (1,882 chains) 0.823 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LigandMPNN on CATH4.3 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LigandMPNN', column 'CATH4.3 test' (page 16 of 21) Configuration: LigandMPNN (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: RecentPDB (121 chains from 111 recent PDB entries) 0.829 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LigandMPNN on RecentPDB (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LigandMPNN', column 'RecentPDB' (page 16 of 21) Configuration: LigandMPNN (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS50 (50 chains) 0.912 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LigandMPNN on TS50 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LigandMPNN', column 'TS50' (page 16 of 21) Configuration: LigandMPNN (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS500 (500 chains) 0.889 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LigandMPNN on TS500 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LigandMPNN', column 'TS500' (page 16 of 21) Configuration: LM-Design 650M (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.2 test (1,120 chains) 0.819 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LM-Design 650M on CATH4.2 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LM-Design 650M', column 'CATH4.2 test' (page 16 of 21) Configuration: LM-Design 650M (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: CATH4.3 test (1,882 chains) 0.814 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LM-Design 650M on CATH4.3 test (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LM-Design 650M', column 'CATH4.3 test' (page 16 of 21) Configuration: LM-Design 650M (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: RecentPDB (121 chains from 111 recent PDB entries) 0.784 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LM-Design 650M on RecentPDB (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LM-Design 650M', column 'RecentPDB' (page 16 of 21) Configuration: LM-Design 650M (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS50 (50 chains) 0.862 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LM-Design 650M on TS50 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LM-Design 650M', column 'TS50' (page 16 of 21) Configuration: LM-Design 650M (HyInv report) Protocol: scTM of designed sequences under ESMFold2 (HyInv report) Dataset: TS500 (500 chains) 0.881 Sctm (Mean TM-score of the predicted structure to the native backbone, averaged over the designs of a target and then over targets (macro-average); computed on the CA trace at valid sites, normalised by the native length and clipped at one; ESMFold2 as the structure predictor, which is also the reinforcement-learning reward of HyInv)unitless · higher
Uncertainty: Not reported by the source
Coverage: Not reported scored / Not reported eligible
Author-reported evaluation · Source checked Evidence concern: excluded from comparisons Appendix C states 'We do not claim zero sequence overlap between the training data and the evaluation panels', and Appendix G's homology screen is a coverage rule under which partial hits remain (Table 10: 28 of 74 de novo backbones and 7 of 12 pocket backbones have partial hits, and all 4 named enzymes are homologous to training data by design). Results on these panels are therefore not clean held-out measurements and must not be compared with results from sources that do establish no overlap.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix C 'Data and pool', final bullet; Appendix G 'Homology of the panels' and Table 10
The ten baselines are single published settings at one temperature each, so their design-set diversity differs from HyInv's: on CATH4.3 test HyInv samples at diversity 0.331 against 0.093 for PiFold, 0.104 for CMLM, 0.144 for LM-Design 650M and 0.000 for DPLM-2 3B, which decodes by argmax. Appendix D.2 states that these four differences are not at matched diversity, and Appendix A states that even for the closest baselines the comparisons are 'close to, but not exactly at, equal diversity', the largest offset of 0.016 being worth at most 0.0011 scTM at the development slope. The scTM and scTM difference results of the protocols named here must not be read as a matched-diversity ranking. The results read at a stated diversity level (Tables 8 and 9 and Appendix D.3) are not covered by this entry, because they are the ones the report reads on a fitted curve at a fixed diversity; nor are the recovery, diversity, pLDDT, scRMSD and lDDT results, whose protocols carry the same single-setting caveat in their own limitations and whose metric qualifiers say the value is at each method's own setting.HyInv Technical Report (Hyra Team, 2026-10-14) · Appendix D.1 and D.2; Table 7; Figure 14 diversity block; Section 5.1 'Baselines' and 'HyInv'
The inference-time results were measured with the supervised-stage weights and a rewrite temperature of 0.23, not with the released checkpoint and its default temperatures, and the quality read-outs of the two numerical paths used the RL checkpoint at update 288, not update 640. Speed and quality in this report are therefore measured on different weights, and the 57x figure compares HyInv's optimised BF16 path with ProteinMPNN's released implementation rather than with an equally optimised baseline.HyInv Technical Report (Hyra Team, 2026-10-14) · Table 1 footnote; Appendix F 'Timing protocol' and 'Baseline timings'
Methods, coverage and source LM-Design 650M on TS500 (scTM of designed sequences under ESMFold2, HyInv report)
hyinv-20261106-protocol-sctm-esmfold2
Aggregation: Not reported
HyInv Technical Report (Hyra Team, 2026-10-14) · Table 6, row 'LM-Design 650M', column 'TS500' (page 16 of 21)