ContraNovo

algorithm · Transformer (AR)

ContraNovo: algorithm · Transformer (AR). Contrastive learning

Contrastive learning

Kind algorithm
Deep learning yes
Acquisition DDA
Family Transformer (AR)

Code

Live stars, open issues and last-push figures are on the Code activity chart.

Checkpoints

Version Trained on Host Licence Size Checked Backup
AAAI release — Google Drive MIT 275 MB verified 2026-10-02 copy

A host marked archival has a DOI and keeps what it is given. The others can move or disappear, which is why they are checked rather than merely listed. verified means the bytes were fetched and hashed on the date shown; live means only that the host answered when last asked. Where a copy is linked, it is a backup of someone else’s weights kept in case the original link goes stale; the original is the link to cite and to prefer.

Data

What the papers describing this method trained it on and reported its results on, as each paper states it. A version is named only where the paper names one.

Trained on

Evaluated on

Benchmarks

  • denovo_benchmarks: median peptide-level average precision 0.808 over 86 datasets, median rank 5 of 15 (version bm-1.0.0).
  • ProteoBench, on the nine-species benchmark, ProteoBench selection: peptide-level AUC 0.883; precision 0.670 at 98% coverage; amino-acid AUC 0.838 (version 1.0, beam search, submitted 2026-08-07).

Both are mass-based matches on the tool’s most recent run. What these numbers mean.

Reported comparisons (2)

The comparison tables this method’s own papers print, standardised: every value on a 0-1 scale, methods down the side, the measure and then the species across. These are numbers papers report about themselves and their baselines. They are not a leaderboard, and they do not compare across tables: each was produced by a different group, on the dataset named in its corner, with each baseline either retrained, run from released weights or quoted from another paper. Where the paper says which, it follows the method’s name (hover it for the sentence); most papers do not say. Bold is the best value in a column and underline the runner-up, our ranking rather than the paper’s own marks.

Table 1

ContraNovo: A Contrastive Learning Approach to Enhance De Novo Peptide Sequencing, page 6: Comparison of the performance of ContraNovo and five baseline methods on 9-species-V1 test set. The bold font indicates the best performance.

Nine-species benchmark
original (DeepNovo, 2017) 		Amino acid precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
PEAKS 0.600 0.639 0.748 0.673 0.633 0.728 0.644 0.719 0.586 0.663
DeepNovo 0.623 0.610 0.750 0.694 0.630 0.731 0.679 0.742 0.602 0.673
PointNovo 0.626 0.606 0.779 0.712 0.644 0.733 0.730 0.768 0.589 0.687
Casanovo 0.689 0.586 0.684 0.679 0.629 0.721 0.668 0.749 0.603 0.667
Casanovo V2 0.760 0.676 0.752 0.755 0.706 0.785 0.748 0.790 0.681 0.739
ContraNovo 0.798 0.771 0.797 0.799 0.745 0.810 0.807 0.828 0.711 0.785
Nine-species benchmark
original (DeepNovo, 2017) 		Peptide precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
PEAKS 0.197 0.277 0.428 0.356 0.287 0.403 0.362 0.387 0.203 0.322
DeepNovo 0.286 0.293 0.462 0.422 0.330 0.454 0.436 0.449 0.253 0.376
PointNovo 0.355 0.351 0.534 0.478 0.396 0.513 0.511 0.518 0.298 0.439
Casanovo 0.426 0.341 0.490 0.478 0.406 0.521 0.506 0.537 0.330 0.448
Casanovo V2 0.483 0.446 0.599 0.557 0.493 0.618 0.589 0.622 0.446 0.539
ContraNovo 0.567 0.622 0.674 0.630 0.576 0.672 0.677 0.688 0.486 0.621

Table2

ContraNovo: A Contrastive Learning Approach to Enhance De Novo Peptide Sequencing, page 7: Comparison of the performance of ContraNovo and CasaNovoV2 methods on 9-species-V2 test set. The bold font indicates the best performance.

Nine-species benchmark
revised (build not stated) 		Amino acid precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
Casanovo V2 0.792 0.836 0.851 0.828 0.759 0.823 0.858 0.852 0.760 0.818
ContraNovo 0.816 0.893 0.875 0.863 0.804 0.860 0.909 0.861 0.778 0.851
Nine-species benchmark
revised (build not stated) 		Peptide precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
Casanovo V2 0.520 0.651 0.729 0.675 0.572 0.692 0.702 0.717 0.505 0.640
ContraNovo 0.566 0.745 0.719 0.716 0.608 0.716 0.791 0.713 0.521 0.677

Compared against in other papers (5)

14 tables from 5 papers that report this method as a baseline, newest first. The caveats of the comparisons above hold twice over: each paper ran, retrained or quoted this method its own way, so its number here is that paper’s claim, not a measurement of the method. Where a paper says which it did, the word follows the method’s name.

ScopeNovo: a De Novo Peptide Sequencing Method Using Spectrum-Spectrum Similarity to Refine the Prediction Accuracy (2026) · 2 tables

Table 1 (Nine-species · revised (build not stated))

ScopeNovo: a De Novo Peptide Sequencing Method Using Spectrum-Spectrum Similarity to Refine the Prediction Accuracy, page 5: Comparison of de novo peptide sequencing methods on 9-species-V1 and 9-species-V2 benchmarks. For ScopeNovo we report two operating regimes: Setting 1 (External Library, spectrum-consistency reranking against a curated external spectral library), which characterizes the upper-bound performance attainable when high-quality external spectral resources are available; and Setting 2 (Self-Constructed Library), in which the reference library is bootstrapped entirely from high-confidence in-batch de novo predictions on the test data itself, placing ScopeNovo under the same resource assumptions as the baselines and yielding a direct comparison. In each column, the best value among the baselines and ScopeNovo (Setting 2) is in bold; ScopeNovo (Setting 1) is reported separately for reference and excluded from the bold comparison.

Nine-species benchmark
revised (build not stated) 		Amino acid precision
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
Casanovo 0.923 0.787 0.776 0.856 0.828 0.806 0.854 0.874 0.865 0.881
Casanovo V2 0.843 0.799 0.788 0.865 0.836 0.813 0.867 0.882 0.875 0.865
InstaNovo 0.852 0.751 0.799 0.847 0.848 0.634 0.894 0.864 0.877 0.850
ContraNovo 0.863 0.802 0.817 0.909 0.864 0.830 0.903 0.881 0.883 0.865
AdaNovo 0.792 0.816 0.756 0.800 0.804 0.757 0.759 0.831 0.801 0.794
RankNovo 0.836 0.836 0.891 0.836 0.879 0.920 0.896 0.893 0.879 0.880
π-PrimeNovo 0.874 0.806 0.824 0.870 0.869 0.808 0.921 0.885 0.891 0.866
ScopeNovo Setting 1 0.885 0.822 0.839 0.884 0.882 0.841 0.931 0.897 0.900 0.885
ScopeNovo Setting 2 0.877 0.814 0.829 0.879 0.874 0.821 0.926 0.889 0.892 0.868
Nine-species benchmark
revised (build not stated) 		Peptide accuracy
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
Casanovo 0.687 0.522 0.579 0.662 0.656 0.522 0.645 0.730 0.738 0.682
Casanovo V2 0.693 0.532 0.587 0.670 0.661 0.531 0.654 0.737 0.743 0.683
InstaNovo 0.704 0.512 0.625 0.670 0.666 0.504 0.679 0.721 0.758 0.698
ContraNovo 0.709 0.533 0.615 0.759 0.698 0.567 0.722 0.729 0.725 0.701
AdaNovo 0.636 0.624 0.597 0.590 0.641 0.472 0.538 0.693 0.653 0.635
RankNovo 0.713 0.603 0.617 0.717 0.712 0.559 0.744 0.724 0.680 0.682
π-PrimeNovo 0.712 0.572 0.612 0.710 0.714 0.548 0.766 0.719 0.684 0.697
ScopeNovo Setting 1 0.748 0.664 0.669 0.762 0.772 0.664 0.816 0.771 0.702 0.737
ScopeNovo Setting 2 0.728 0.619 0.641 0.750 0.743 0.611 0.788 0.745 0.691 0.716

The paper bolds the best of the baselines and ScopeNovo (Setting 2) only: 'ScopeNovo (Setting 1) is reported separately for reference and excluded from the bold comparison.' Bold and underline here rank every row, Setting 1 included; the paper's own marks were checked against its rule, and match it.

Table 1 (Nine-species · original (DeepNovo, 2017))

ScopeNovo: a De Novo Peptide Sequencing Method Using Spectrum-Spectrum Similarity to Refine the Prediction Accuracy, page 5: Comparison of de novo peptide sequencing methods on 9-species-V1 and 9-species-V2 benchmarks. For ScopeNovo we report two operating regimes: Setting 1 (External Library, spectrum-consistency reranking against a curated external spectral library), which characterizes the upper-bound performance attainable when high-quality external spectral resources are available; and Setting 2 (Self-Constructed Library), in which the reference library is bootstrapped entirely from high-confidence in-batch de novo predictions on the test data itself, placing ScopeNovo under the same resource assumptions as the baselines and yielding a direct comparison. In each column, the best value among the baselines and ScopeNovo (Setting 2) is in bold; ScopeNovo (Setting 1) is reported separately for reference and excluded from the bold comparison.

Nine-species benchmark
original (DeepNovo, 2017) 		Amino acid precision
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
Casanovo 0.788 0.676 0.708 0.676 0.754 0.765 0.761 0.792 0.756 0.743
Casanovo V2 0.801 0.687 0.721 0.687 0.767 0.774 0.768 0.802 0.768 0.754
InstaNovo 0.802 0.656 0.718 0.691 0.756 0.742 0.794 0.774 0.767 0.745
ContraNovo 0.828 0.700 0.764 0.778 0.798 0.803 0.804 0.812 0.787 0.785
AdaNovo 0.858 0.798 0.767 0.690 0.804 0.778 0.878 0.818 0.791 0.798
RankNovo 0.807 0.771 0.880 0.784 0.838 0.823 0.876 0.846 0.828 0.827
π-PrimeNovo 0.840 0.715 0.767 0.725 0.799 0.797 0.823 0.822 0.795 0.789
ScopeNovo Setting 1 0.849 0.715 0.778 0.729 0.811 0.809 0.816 0.825 0.796 0.797
ScopeNovo Setting 2 0.844 0.722 0.774 0.732 0.806 0.809 0.829 0.828 0.798 0.795
Nine-species benchmark
original (DeepNovo, 2017) 		Peptide accuracy
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
Casanovo 0.621 0.437 0.494 0.446 0.568 0.485 0.593 0.622 0.601 0.551
Casanovo V2 0.630 0.445 0.504 0.455 0.579 0.493 0.600 0.629 0.607 0.559
InstaNovo 0.652 0.459 0.535 0.515 0.595 0.525 0.635 0.635 0.618 0.582
ContraNovo 0.688 0.486 0.580 0.625 0.633 0.574 0.678 0.674 0.673 0.628
AdaNovo 0.753 0.638 0.620 0.494 0.671 0.546 0.797 0.684 0.660 0.660
RankNovo 0.765 0.592 0.642 0.621 0.688 0.595 0.765 0.720 0.726 0.687
π-PrimeNovo 0.725 0.532 0.602 0.575 0.654 0.572 0.701 0.701 0.703 0.650
ScopeNovo Setting 1 0.789 0.610 0.665 0.628 0.731 0.660 0.755 0.752 0.744 0.711
ScopeNovo Setting 2 0.749 0.572 0.643 0.624 0.690 0.637 0.733 0.732 0.721 0.684

The paper bolds the best of the baselines and ScopeNovo (Setting 2) only: 'ScopeNovo (Setting 1) is reported separately for reference and excluded from the bold comparison.' Bold and underline here rank every row, Setting 1 included; the paper's own marks were checked against its rule, and match it.

MassNet: billion-scale AI-ready mass spectral corpus empowering scalable deep learning in proteomics (2026) · 4 tables

Fig. 4b

MassNet: billion-scale AI-ready mass spectral corpus empowering scalable deep learning in proteomics, page 7: Comparison of XuanjiNovo and existing de novo peptide sequencing models on the 15-species benchmark dataset. b, Cross-species peptide recall aggregated by species, illustrating overall performance differences across taxa.

15-species benchmark
Nature Methods (307 files) 		Peptide recall
Acinetobacter calcoaceticus Aedes aegypti Arabidopsis thaliana Aspergillus fumigatus Bacteroides thetaiotaomicron Drosophila melanogaster Escherichia coli Homo sapiens Mus musculus Pyrococcus furiosus Saccharomyces cerevisiae Toxoplasma gondii
π-HelixNovo · released 0.41 0.403 0.183 0.286 0.413 0.188 0.114 0.184 0.307 0.398 0.386 0.186
Casanovo V2 · released 0.6 0.57 0.266 0.395 0.652 0.319 0.329 0.224 0.458 0.516 0.598 0.262
InstaNovo+ · released 0.632 0.606 0.262 0.412 0.661 0.346 0.345 0.259 0.524 0.545 0.6 0.241
ContraNovo 0.654 0.615 0.283 0.436 0.678 0.373 0.368 0.299 0.54 0.544 0.626 0.277
XuanjiNovo 0.779 0.757 0.333 0.582 0.793 0.444 0.476 0.449 0.676 0.668 0.691 0.355
π-PrimeNovo 0.733 0.686 0.323 0.51 0.78 0.429 0.466 0.324 0.604 0.616 0.692 0.308
15-species benchmark
Nature Methods (307 files) 		Peptide recall
Trypanosoma brucei Zea mays Solanum lycopersicum
π-HelixNovo · released 0.348 0.443 0.001
Casanovo V2 · released 0.588 0.557 0.011
InstaNovo+ · released 0.623 0.649 0.004
ContraNovo 0.641 0.653 0.062
XuanjiNovo 0.762 0.799 0.252
π-PrimeNovo 0.738 0.731 0.06

Read from the paper's source data: every value is a cell of 'Source Data.xlsx', sheet Figure4B, behind a figure that prints no values. Nothing was estimated from bar heights. Values from the paper's Source Data file (sheet Figure4B); the figure prints none. XuanjiNovo is XuanjiNovo-100M, per the caption. THE TEXT AND THE SHEET DISAGREE once: p. 5 says the other models on S. lycopersicum ranged 'from as low as 0.01 (π-HelixNovo)', the sheet gives 0.001.

Supplementary Fig. 3c

MassNet: billion-scale AI-ready mass spectral corpus empowering scalable deep learning in proteomics, page 5: Evaluation of de novo peptide sequencing performance on other benchmark datasets. (A) Peptide recall across nine species using seven models: InstaNovo+, π-HelixNovo, ContraNovo, CasaNovo V2, π-PrimeNovo, XuanjiNovo-100M, and XuanjiNovo-130M. (B) Peptide recall on the HCC dataset. (C) Peptide recall on the PT dataset. The π-HelixNovo model is publicly available and was trained on datasets derived from eight species, excluding human. XuanjiNovo-100M: base model; XuanjiNovo-130M: fine-tuned on 30M PSMs from MassIVE.

ProteomeTools Peptide recall
PT
Casanovo V2 0.459
InstaNovo+ 0.479
ContraNovo 0.574
π-PrimeNovo 0.586
XuanjiNovo 0.605

Read from the paper's source data: every value is a cell of 'Supplementary Data 1.xlsx', sheet SFigure3C, behind a figure that prints no values. Nothing was estimated from bar heights. Values from Supplementary Data 1 (sheet SFigure3C); the figure prints none. The figure draws six bars and the sheet has five: π-HelixNovo (about 0.26 by its bar) has no row, so it is not recorded. THE TEXT AND THE SHEET DISAGREE: p. 8 says XuanjiNovo-100M 'achieved a recall of 0.62' on PT, the sheet gives 0.605.

Supplementary Fig. 3b

MassNet: billion-scale AI-ready mass spectral corpus empowering scalable deep learning in proteomics, page 5: Evaluation of de novo peptide sequencing performance on other benchmark datasets. (A) Peptide recall across nine species using seven models: InstaNovo+, π-HelixNovo, ContraNovo, CasaNovo V2, π-PrimeNovo, XuanjiNovo-100M, and XuanjiNovo-130M. (B) Peptide recall on the HCC dataset. (C) Peptide recall on the PT dataset. The π-HelixNovo model is publicly available and was trained on datasets derived from eight species, excluding human. XuanjiNovo-100M: base model; XuanjiNovo-130M: fine-tuned on 30M PSMs from MassIVE.

Proteomics identifies new therapeutic targets of early-stage hepatocellular carcinoma
as deposited 		Peptide recall
HCC
Casanovo V2 0.168
InstaNovo+ 0.087
ContraNovo 0.373
π-PrimeNovo 0.382
XuanjiNovo 0.676

Read from the paper's source data: every value is a cell of 'Supplementary Data 1.xlsx', sheet SFigure3B, behind a figure that prints no values. Nothing was estimated from bar heights. Values from Supplementary Data 1 (sheet SFigure3B); the figure prints none. The figure draws six bars and the sheet has five: π-HelixNovo, whose bar is barely above zero, has no row, so it is not recorded rather than read off the bar. XuanjiNovo is XuanjiNovo-100M, per the caption of Fig. 4.

Supplementary Fig. 3a

MassNet: billion-scale AI-ready mass spectral corpus empowering scalable deep learning in proteomics, page 5: Evaluation of de novo peptide sequencing performance on other benchmark datasets. (A) Peptide recall across nine species using seven models: InstaNovo+, π-HelixNovo, ContraNovo, CasaNovo V2, π-PrimeNovo, XuanjiNovo-100M, and XuanjiNovo-130M. (B) Peptide recall on the HCC dataset. (C) Peptide recall on the PT dataset. The π-HelixNovo model is publicly available and was trained on datasets derived from eight species, excluding human. XuanjiNovo-100M: base model; XuanjiNovo-130M: fine-tuned on 30M PSMs from MassIVE.

Nine-species benchmark
original (DeepNovo, 2017) 		Peptide recall
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae
π-HelixNovo · released 0.596 0.388 0.473 0.392 0.56 0.483 0.623 0.56 0.568
InstaNovo+ 0.576 0.417 0.467 0.456 0.528 0.471 0.565 0.585 0.56
Casanovo V2 0.622 0.446 0.499 0.455 0.565 0.486 0.594 0.624 0.598
ContraNovo 0.688 0.487 0.576 0.622 0.63 0.566 0.677 0.672 0.674
π-PrimeNovo 0.721 0.531 0.603 0.574 0.65 0.567 0.702 0.697 0.697
XuanjiNovo 100M 0.43 0.393 0.435 0.589 0.468 0.502 0.499 0.536 0.407
XuanjiNovo 130M 0.689 0.503 0.574 0.6 0.628 0.682 0.595 0.681 0.67

Read from the paper's source data: every value is a cell of 'Supplementary Data 1.xlsx', sheet SFigure3A, behind a figure that prints no values. Nothing was estimated from bar heights. Values from the paper's Supplementary Data 1 (sheet SFigure3A); the figure prints none. The sheet names InstaNovo where the legend draws InstaNovo+. THE TEXT AND THE SHEET DISAGREE once: p. 8 puts XuanjiNovo-130M on M. musculus at 'approximately 0.67', the sheet at 0.682.

Bidirectional Representations Augmented Autoregressive Biological Sequence Generation (2025) · 4 tables

Table7

Bidirectional Representations Augmented Autoregressive Biological Sequence Generation, page 26: Comparison of the performance of CROSSNOVO and baseline methods on IgG1-Human. Bold values indicate the best performance.

The same table is also printed in Bidirectional Representations Augmented Autoregressive Biological Sequence Generation (Table7, page 26).

Monoclonal antibody de novo assembly
as deposited 		Amino acid precision
HC AspN HC Chymotrypsin HC GluC HC LysC HC Proteinase K HC Trypsin LC AspN LC LysC Average
Casanovo V2 0.520 0.472 0.605 0.757 0.354 0.759 0.666 0.778 0.642
ContraNovo 0.580 0.565 0.642 0.790 0.348 0.787 0.702 0.793 0.676
CrossNovo 0.613 0.617 0.694 0.814 0.367 0.803 0.719 0.807 0.702
Monoclonal antibody de novo assembly
as deposited 		Peptide recall
HC AspN HC Chymotrypsin HC GluC HC LysC HC Proteinase K HC Trypsin LC AspN LC LysC Average
Casanovo V2 0.265 0.274 0.399 0.569 0.206 0.595 0.325 0.625 0.446
ContraNovo 0.396 0.372 0.437 0.653 0.274 0.675 0.499 0.646 0.529
CrossNovo 0.415 0.421 0.512 0.701 0.275 0.699 0.544 0.676 0.560

Table6

Bidirectional Representations Augmented Autoregressive Biological Sequence Generation, page 26: Comparison of the performance of CROSSNOVO and baseline methods on WIgG1-Mouse. The bold font indicates the best performance.

The same table is also printed in Bidirectional Representations Augmented Autoregressive Biological Sequence Generation (Table6, page 26).

Monoclonal antibody de novo assembly
as deposited 		Amino acid precision Peptide recall
HC AspN HC Chymotrypsin HC Trypsin LC AspN Average HC AspN HC Chymotrypsin HC Trypsin LC AspN Average
Casanovo V2 0.714 0.591 0.723 0.668 0.674 0.557 0.483 0.636 0.456 0.533
ContraNovo 0.750 0.612 0.650 0.649 0.665 0.649 0.545 0.671 0.519 0.596
CrossNovo 0.769 0.640 0.747 0.724 0.720 0.662 0.577 0.699 0.581 0.630

Table2

Bidirectional Representations Augmented Autoregressive Biological Sequence Generation, page 8: Comparison of the performance of CROSSNOVO and baseline methods on 9-species-v2 test set. The bold font indicates the best performance.

The same table is also printed in Bidirectional Representations Augmented Autoregressive Biological Sequence Generation (Table2, page 8), Distilling Non-Autoregressive Model Knowledge for Autoregressive De Novo Peptide Sequencing (Table2, page 8).

Nine-species benchmark
revised (build not stated) 		Amino acid precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
π-PrimeNovo 0.839 0.893 0.932 0.908 0.862 0.909 0.931 0.921 0.827 0.891
Casanovo V2 0.813 0.872 0.915 0.877 0.823 0.891 0.891 0.888 0.791 0.862
ContraNovo 0.839 0.920 0.919 0.896 0.848 0.898 0.913 0.901 0.807 0.882
CrossNovo 0.857 0.937 0.939 0.920 0.880 0.914 0.939 0.927 0.837 0.906
Nine-species benchmark
revised (build not stated) 		Peptide recall
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
π-PrimeNovo 0.627 0.795 0.884 0.812 0.742 0.824 0.837 0.849 0.626 0.777
Casanovo V2 0.555 0.712 0.837 0.754 0.669 0.783 0.772 0.793 0.558 0.714
ContraNovo 0.616 0.820 0.854 0.780 0.711 0.794 0.799 0.815 0.575 0.752
CrossNovo 0.651 0.850 0.885 0.819 0.751 0.816 0.847 0.850 0.607 0.786

Table1

Bidirectional Representations Augmented Autoregressive Biological Sequence Generation, page 8: Comparison of the performance of CROSSNOVO and baseline methods on the 9-species-v1 test set.

The same table is also printed in Bidirectional Representations Augmented Autoregressive Biological Sequence Generation (Table1, page 8).

Nine-species benchmark
original (DeepNovo, 2017) 		Amino acid precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
PEAKS 0.600 0.639 0.748 0.673 0.633 0.728 0.644 0.719 0.586 0.663
π-PrimeNovo 0.784 0.729 0.802 0.801 0.763 0.815 0.822 0.846 0.734 0.788
DeepNovo 0.623 0.610 0.750 0.694 0.630 0.731 0.679 0.742 0.602 0.673
PointNovo 0.626 0.606 0.779 0.712 0.644 0.733 0.730 0.768 0.589 0.687
Casanovo 0.689 0.586 0.684 0.679 0.629 0.721 0.668 0.749 0.603 0.667
InstaNovo 0.703 0.636 0.691 0.712 0.660 0.732 0.711 0.739 0.619 0.689
Casanovo V2 0.760 0.676 0.752 0.755 0.706 0.785 0.748 0.790 0.681 0.739
π-HelixNovo 0.765 0.665 0.768 0.784 0.757 0.721 0.793 0.816 0.681 0.750
ContraNovo 0.798 0.771 0.797 0.799 0.745 0.810 0.807 0.828 0.711 0.785
CrossNovo 0.816 0.800 0.814 0.826 0.785 0.830 0.831 0.856 0.740 0.811
Nine-species benchmark
original (DeepNovo, 2017) 		Peptide recall
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
PEAKS 0.197 0.277 0.428 0.356 0.287 0.403 0.362 0.387 0.203 0.322
π-PrimeNovo 0.567 0.574 0.697 0.650 0.603 0.697 0.702 0.721 0.531 0.638
DeepNovo 0.286 0.293 0.462 0.422 0.330 0.454 0.436 0.449 0.253 0.376
PointNovo 0.355 0.351 0.534 0.478 0.396 0.513 0.511 0.518 0.298 0.439
Casanovo 0.426 0.341 0.490 0.478 0.406 0.521 0.506 0.537 0.330 0.448
InstaNovo 0.471 0.455 0.559 0.528 0.466 0.732 0.564 0.576 0.416 0.530
Casanovo V2 0.483 0.446 0.599 0.557 0.493 0.618 0.589 0.622 0.446 0.539
π-HelixNovo 0.483 0.392 0.568 0.560 0.473 0.560 0.623 0.596 0.388 0.517
ContraNovo 0.567 0.622 0.674 0.630 0.576 0.672 0.677 0.688 0.486 0.621
CrossNovo 0.596 0.661 0.698 0.660 0.610 0.695 0.716 0.726 0.518 0.654

In the paper: column peptide recall, Mouse: the original table only underlined ContraNovo (Contra.) (0.567); π-PrimeNovo (Prime.) (0.567) ties with it and is underlined here too.
In the paper: column peptide recall, Tomato: the original table bolded π-PrimeNovo (Prime.) (0.697), InstaNovo (Insta) (0.732) and underlined CrossNovo (Ours) (0.695).

Universal Biological Sequence Reranking for Improved De Novo Peptide Sequencing (2025) · 3 tables

Table7

Universal Biological Sequence Reranking for Improved De Novo Peptide Sequencing, page 17: Evaluation of RankNovo in comparison to baseline and base methods on the 9-species-V2 test set. Bolded entries indicate the best-performing models. The symbol “†” indicates that the model serves as both a baseline and a base model. “⋆” signifies that the base models were developed and trained by us.

The same table is also printed in RankNovo: A Universal Reranking Approach for Robust De Novo Peptide Sequencing (Table2, page 8).

Nine-species benchmark
revised (build not stated) 		Amino acid precision
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
Casanovo V2 · released 0.888 0.791 0.823 0.872 0.877 0.813 0.891 0.891 0.915 0.862
ContraNovo · released 0.901 0.807 0.848 0.920 0.896 0.839 0.913 0.898 0.919 0.882
ByNovo 0.92 0.823 0.876 0.917 0.914 0.841 0.932 0.912 0.934 0.897
Casanovo reverse decoding · retrained 0.876 0.804 0.814 0.891 0.867 0.821 0.881 0.891 0.898 0.860
ContraNovo reverse decoding · retrained 0.909 0.815 0.865 0.923 0.901 0.849 0.919 0.907 0.925 0.890
ByNovo reverse decoding 0.919 0.822 0.879 0.912 0.912 0.843 0.932 0.913 0.936 0.897
RankNovo 0.926 0.838 0.885 0.929 0.920 0.860 0.938 0.918 0.938 0.906
Nine-species benchmark
revised (build not stated) 		Peptide recall
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
Casanovo V2 · released 0.793 0.558 0.669 0.712 0.754 0.555 0.772 0.783 0.837 0.714
ContraNovo · released 0.815 0.575 0.711 0.820 0.780 0.616 0.799 0.794 0.854 0.752
ByNovo 0.833 0.582 0.731 0.789 0.799 0.596 0.814 0.807 0.871 0.758
Casanovo reverse decoding · retrained 0.759 0.558 0.643 0.732 0.723 0.558 0.721 0.768 0.799 0.696
ContraNovo reverse decoding · retrained 0.821 0.581 0.719 0.815 0.779 0.620 0.804 0.798 0.861 0.755
ByNovo reverse decoding 0.831 0.585 0.729 0.781 0.794 0.589 0.815 0.803 0.873 0.756
RankNovo 0.851 0.620 0.752 0.820 0.813 0.629 0.836 0.822 0.885 0.781

RankNovo reranks the candidates of six base models. Casanovo V2 and ContraNovo are the released models; ByNovo is RankNovo's own in-house model, with an auxiliary b/y ion-annotation head (its Appendix B.4); R-Casanovo, R-ContraNovo and R-ByNovo are reverse-decoding versions of the three, trained by RankNovo's authors and recorded as the 'reverse decoding' variant. All six were trained on MassIVE-KB and evaluated zero-shot. '†' marks a model that is both baseline and base model, '⋆' one the authors trained. The plain Casanovo row repeats the values other papers print for Casanovo (0.749 and 0.537 on Bacillus) and the paper does not say where they come from.
In the paper: column peptide recall, Human: the original table only bolded RankNovo (0.820); ContraNovo (ContraNovo†) (0.820) ties with it and is bolded here too.

Table6

Universal Biological Sequence Reranking for Improved De Novo Peptide Sequencing, page 17: Evaluation of RankNovo in comparison to baseline and base methods on the 9-species-V1 test set. Bolded entries indicate the best-performing models. The symbol “†” indicates that the model serves as both a baseline and a base model. “⋆” signifies that the base models were developed and trained by us.

The same table is also printed in RankNovo: A Universal Reranking Approach for Robust De Novo Peptide Sequencing (Table1, page 8).

Nine-species benchmark
original (DeepNovo, 2017) 		Amino acid precision
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
PEAKS 0.719 0.586 0.633 0.639 0.673 0.600 0.644 0.728 0.748 0.663
DeepNovo 0.742 0.602 0.630 0.610 0.694 0.623 0.679 0.731 0.750 0.673
PointNovo 0.768 0.589 0.644 0.606 0.712 0.626 0.730 0.733 0.779 0.687
Casanovo 0.749 0.603 0.629 0.586 0.679 0.689 0.668 0.721 0.684 0.667
Casanovo V2 · released 0.806 0.685 0.727 0.69 0.774 0.768 0.769 0.799 0.762 0.753
ContraNovo · released 0.828 0.706 0.761 0.771 0.798 0.799 0.804 0.808 0.782 0.784
ByNovo 0.858 0.723 0.791 0.767 0.823 0.803 0.836 0.828 0.804 0.804
Casanovo reverse decoding · retrained 0.804 0.699 0.728 0.719 0.769 0.776 0.782 0.795 0.762 0.759
ContraNovo reverse decoding · retrained 0.839 0.716 0.775 0.782 0.806 0.811 0.816 0.822 0.798 0.796
ByNovo reverse decoding 0.855 0.724 0.794 0.762 0.821 0.81 0.835 0.831 0.762 0.799
RankNovo 0.874 0.746 0.81 0.802 0.84 0.828 0.859 0.844 0.816 0.824
Nine-species benchmark
original (DeepNovo, 2017) 		Peptide recall
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
PEAKS 0.387 0.203 0.287 0.277 0.356 0.197 0.362 0.403 0.428 0.322
DeepNovo 0.449 0.253 0.330 0.293 0.422 0.286 0.436 0.454 0.462 0.376
PointNovo 0.518 0.298 0.396 0.351 0.478 0.355 0.511 0.513 0.534 0.439
Casanovo 0.537 0.330 0.406 0.341 0.478 0.426 0.506 0.521 0.490 0.448
Casanovo V2 · released 0.646 0.46 0.527 0.492 0.592 0.493 0.628 0.637 0.629 0.567
ContraNovo · released 0.684 0.487 0.576 0.624 0.628 0.563 0.676 0.655 0.669 0.618
ByNovo 0.708 0.499 0.597 0.584 0.639 0.545 0.696 0.667 0.676 0.623
Casanovo reverse decoding · retrained 0.628 0.467 0.515 0.511 0.57 0.505 0.611 0.611 0.601 0.558
ContraNovo reverse decoding · retrained 0.682 0.499 0.583 0.606 0.621 0.566 0.673 0.654 0.664 0.616
ByNovo reverse decoding 0.703 0.493 0.59 0.554 0.637 0.543 0.685 0.659 0.629 0.610
RankNovo 0.738 0.539 0.63 0.642 0.672 0.583 0.733 0.691 0.703 0.660

RankNovo reranks the candidates of six base models. Casanovo V2 and ContraNovo are the released models; ByNovo is RankNovo's own in-house model, with an auxiliary b/y ion-annotation head (its Appendix B.4); R-Casanovo, R-ContraNovo and R-ByNovo are reverse-decoding versions of the three, trained by RankNovo's authors and recorded as the 'reverse decoding' variant. All six were trained on MassIVE-KB and evaluated zero-shot. '†' marks a model that is both baseline and base model, '⋆' one the authors trained. The plain Casanovo row repeats the values other papers print for Casanovo (0.749 and 0.537 on Bacillus) and the paper does not say where they come from. THE TEXT AND THE TABLE DISAGREE on V1: the prose gives RankNovo an amino-acid precision of 0.829 and a peptide recall '6.1%' above ByNovo; the table prints 0.824, and 0.660 against ByNovo's 0.623. The table's values are recorded.

Table1

Universal Biological Sequence Reranking for Improved De Novo Peptide Sequencing, page 7: Evaluation of RankNovo in comparison to baseline and base models on the 9-species-V1 and 9-species-V2 test sets. Bolded entries indicate the best-performing models. The symbol “†” indicates that the model serves as both a baseline and a base model. “⋆” signifies that the base models were developed and trained by us. Here, ByNovo is the best base model. So the performance of three self-trained base models (R-Casa, R-Contra, R-By) is only provided in Appendix D.1.

Nine-species benchmark Amino acid precision
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
PEAKS 0.719 0.586 0.633 0.639 0.673 0.600 0.644 0.728 0.748 0.663
DeepNovo 0.742 0.602 0.630 0.610 0.694 0.623 0.679 0.731 0.750 0.673
PointNovo 0.768 0.589 0.644 0.606 0.712 0.626 0.730 0.733 0.779 0.687
Casanovo 0.749 0.603 0.629 0.586 0.679 0.689 0.668 0.721 0.684 0.667
Casanovo V2 · released 0.888 0.791 0.823 0.872 0.877 0.813 0.891 0.891 0.915 0.862
ContraNovo · released 0.901 0.807 0.848 0.920 0.896 0.839 0.913 0.898 0.919 0.882
ByNovo 0.920 0.823 0.876 0.917 0.914 0.841 0.932 0.912 0.934 0.897
RankNovo 0.926 0.838 0.885 0.929 0.920 0.860 0.938 0.918 0.938 0.906
Nine-species benchmark Peptide recall
Bacillus subtilis Candidatus Thiodiazotropha endoloripes Apis mellifera Homo sapiens Methanosarcina mazei Mus musculus Vigna mungo Solanum lycopersicum Saccharomyces cerevisiae Average
PEAKS 0.387 0.203 0.287 0.277 0.356 0.197 0.362 0.403 0.428 0.322
DeepNovo 0.449 0.253 0.330 0.293 0.422 0.286 0.436 0.454 0.462 0.376
PointNovo 0.518 0.298 0.396 0.351 0.478 0.355 0.511 0.513 0.534 0.439
Casanovo 0.537 0.330 0.406 0.341 0.478 0.426 0.506 0.521 0.490 0.448
Casanovo V2 · released 0.793 0.558 0.669 0.712 0.754 0.555 0.772 0.783 0.837 0.714
ContraNovo · released 0.815 0.575 0.711 0.820 0.780 0.616 0.799 0.794 0.854 0.752
ByNovo 0.833 0.582 0.731 0.789 0.799 0.596 0.814 0.807 0.871 0.758
RankNovo 0.851 0.620 0.752 0.820 0.813 0.629 0.836 0.822 0.885 0.781

RankNovo reranks the candidates of six base models. Casanovo V2 and ContraNovo are the released models; ByNovo is RankNovo's own in-house model, with an auxiliary b/y ion-annotation head (its Appendix B.4); R-Casanovo, R-ContraNovo and R-ByNovo are reverse-decoding versions of the three, trained by RankNovo's authors and recorded as the 'reverse decoding' variant. All six were trained on MassIVE-KB and evaluated zero-shot. '†' marks a model that is both baseline and base model, '⋆' one the authors trained. The plain Casanovo row repeats the values other papers print for Casanovo (0.749 and 0.537 on Bacillus) and the paper does not say where they come from. THE TEXT AND THE TABLE DISAGREE on V1: the prose gives RankNovo an amino-acid precision of 0.829 and a peptide recall '6.1%' above ByNovo; the table prints 0.824, and 0.660 against ByNovo's 0.623. The table's values are recorded.
In the paper: column amino acid precision, Average: the original table bolded RankNovo (0.824), RankNovo (0.906).
In the paper: column amino acid precision, Bacillus: the original table bolded RankNovo (0.874), RankNovo (0.926).
In the paper: column amino acid precision, C. bacteria: the original table bolded RankNovo (0.746), RankNovo (0.838).
In the paper: column amino acid precision, Honeybee: the original table bolded RankNovo (0.810), RankNovo (0.885).
In the paper: column amino acid precision, Human: the original table bolded RankNovo (0.802), RankNovo (0.929).
In the paper: column amino acid precision, M. mazei: the original table bolded RankNovo (0.840), RankNovo (0.920).
In the paper: column amino acid precision, Mouse: the original table bolded RankNovo (0.828), RankNovo (0.860).
In the paper: column amino acid precision, Ricebean: the original table bolded RankNovo (0.859), RankNovo (0.938).
In the paper: column amino acid precision, Tomato: the original table bolded RankNovo (0.844), RankNovo (0.918).
In the paper: column amino acid precision, Yeast: the original table bolded RankNovo (0.816), RankNovo (0.938).
In the paper: column peptide recall, Average: the original table bolded RankNovo (0.660), RankNovo (0.781).
In the paper: column peptide recall, Bacillus: the original table bolded RankNovo (0.738), RankNovo (0.851).
In the paper: column peptide recall, C. bacteria: the original table bolded RankNovo (0.539), RankNovo (0.620).
In the paper: column peptide recall, Honeybee: the original table bolded RankNovo (0.630), RankNovo (0.752).
In the paper: column peptide recall, Human: the original table bolded RankNovo (0.642), RankNovo (0.820).
In the paper: column peptide recall, M. mazei: the original table bolded RankNovo (0.672), RankNovo (0.813).
In the paper: column peptide recall, Mouse: the original table bolded RankNovo (0.583), RankNovo (0.629).
In the paper: column peptide recall, Ricebean: the original table bolded RankNovo (0.733), RankNovo (0.836).
In the paper: column peptide recall, Tomato: the original table bolded RankNovo (0.691), RankNovo (0.822).
In the paper: column peptide recall, Yeast: the original table bolded RankNovo (0.703), RankNovo (0.885).

Distilling Non-Autoregressive Model Knowledge for Autoregressive De Novo Peptide Sequencing (2025) · 1 table

Table1

Distilling Non-Autoregressive Model Knowledge for Autoregressive De Novo Peptide Sequencing, page 7: Comparison of the performance of CrossNovo and baseline methods on 9-species-v1 test set. The bold font indicates the best performance.

Nine-species benchmark
original (DeepNovo, 2017) 		Amino acid precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
PEAKS 0.600 0.639 0.748 0.673 0.633 0.728 0.644 0.719 0.586 0.663
π-PrimeNovo 0.784 0.729 0.802 0.801 0.763 0.815 0.822 0.846 0.734 0.788
DeepNovo 0.623 0.610 0.750 0.694 0.630 0.731 0.679 0.742 0.602 0.673
PointNovo 0.626 0.606 0.779 0.712 0.644 0.733 0.730 0.768 0.589 0.687
Casanovo 0.689 0.586 0.684 0.679 0.629 0.721 0.668 0.749 0.603 0.667
Casanovo V2 0.760 0.676 0.752 0.755 0.706 0.785 0.748 0.790 0.681 0.739
ContraNovo 0.798 0.771 0.797 0.799 0.745 0.810 0.807 0.828 0.711 0.785
CrossNovo 0.816 0.800 0.814 0.826 0.785 0.830 0.831 0.856 0.740 0.811
Nine-species benchmark
original (DeepNovo, 2017) 		Peptide recall
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes Average
PEAKS 0.197 0.277 0.428 0.356 0.287 0.403 0.362 0.387 0.203 0.322
π-PrimeNovo 0.567 0.574 0.697 0.650 0.603 0.697 0.702 0.721 0.531 0.638
DeepNovo 0.286 0.293 0.462 0.422 0.330 0.454 0.436 0.449 0.253 0.376
PointNovo 0.355 0.351 0.534 0.478 0.396 0.513 0.511 0.518 0.298 0.439
Casanovo 0.426 0.341 0.490 0.478 0.406 0.521 0.506 0.537 0.330 0.448
Casanovo V2 0.483 0.446 0.599 0.557 0.493 0.618 0.589 0.622 0.446 0.539
ContraNovo 0.567 0.622 0.674 0.630 0.576 0.672 0.677 0.688 0.486 0.621
CrossNovo 0.596 0.661 0.698 0.660 0.610 0.695 0.716 0.726 0.518 0.654

In the paper: column peptide recall, Mouse: the original table only underlined ContraNovo (Contra.) (0.567); π-PrimeNovo (Prime.) (0.567) ties with it and is underlined here too.

Paper describing it

Paper using it

An application or evaluation that ran this method. It is not counted among its authors below.

Authors (9)

Zhi Jin, Sheng Xu, Xiang Zhang (Shanghai AI Lab), Tianze Ling, Nanqing Dong, Wanli Ouyang, Zhiqiang Gao, Cheng Chang, Siqi Sun

Seen in the charts

Back to the full map

Back to top