AdaNovo

algorithm · Transformer (AR)

AdaNovo: algorithm · Transformer (AR). Mutual info for PTMs

Mutual info for PTMs

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.

Benchmarks

  • denovo_benchmarks: median peptide-level average precision 0.535 over 86 datasets, median rank 10 of 14 (version bm-1.0.0).
  • ProteoBench, on the nine-species benchmark, ProteoBench selection: peptide-level AUC 0.837; precision 0.589 at 98% coverage; amino-acid AUC 0.882 (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 (5)

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.

Table1

AdaNovo: Adaptive De Novo Peptide Sequencing with Conditional Mutual Information, page 7: Empirical comparison of de novo sequencing models. The table lists the peptide-level and amino acid-level precision of three competing models on all nine benchmark cross-validation folds. Each fold’s test set contains spectra from a single species. Kindly note that peptide-level performance measures are the primary quantifier of the model’s practical utility because the goal is to assign a complete peptide sequence to each spectrum. The best and the second best results are highlighted bold and underlined, respectively.

Nine-species benchmark Peptide precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes
DeepNovo · quoted 0.286 0.293 0.462 0.422 0.330 0.454 0.436 0.449 0.253
PointNovo · quoted 0.355 0.351 0.534 0.478 0.396 0.513 0.511 0.518 0.298
Casanovo · retrained 0.449 0.343 0.568 0.474 0.422 0.463 0.549 0.513 0.347
AdaNovo 0.467 0.373 0.593 0.496 0.431 0.530 0.546 0.528 0.372
Nine-species benchmark Amino acid precision
Mus musculus Homo sapiens Saccharomyces cerevisiae Methanosarcina mazei Apis mellifera Solanum lycopersicum Vigna mungo Bacillus subtilis Candidatus Thiodiazotropha endoloripes
DeepNovo · quoted 0.623 0.610 0.750 0.694 0.630 0.731 0.679 0.742 0.602
PointNovo · quoted 0.626 0.606 0.779 0.712 0.644 0.733 0.730 0.768 0.589
Casanovo · retrained 0.612 0.585 0.753 0.686 0.640 0.720 0.727 0.718 0.617
AdaNovo 0.646 0.618 0.793 0.728 0.650 0.740 0.719 0.739 0.642

Table2

AdaNovo: Adaptive De Novo Peptide Sequencing with Conditional Mutual Information, page 7: Empirical comparison of de novo sequencing models in terms of identifying amino acids with PTMs. The best and the second best results are highlighted bold and underlined, respectively.

Nine-species benchmark Ptm precision
Homo sapiens Vigna mungo Candidatus Thiodiazotropha endoloripes Bacillus subtilis
DeepNovo · quoted 0.369 0.644 0.510 0.483
PointNovo · quoted 0.415 0.653 0.526 0.524
Casanovo · retrained 0.398 0.646 0.508 0.470
AdaNovo 0.483 0.689 0.575 0.565

Table3

AdaNovo: Adaptive De Novo Peptide Sequencing with Conditional Mutual Information, page 8: Ablations on amino acid-level (AA-level) and peptidelevel adaptive training strategies. The results are for the Human test set, which is one of 9-species benchmark (Tran et al., 2017).

Nine-species benchmark Amino acid precision Peptide precision Ptm precision
Homo sapiens Homo sapiens Homo sapiens
Casanovo · retrained 0.585 0.343 0.300
AdaNovo w/oPSM-levelMI 0.607 0.360 0.478
AdaNovo w/oAA-levelCMI 0.594 0.349 0.314
AdaNovo 0.618 0.373 0.483

Table4

AdaNovo: Adaptive De Novo Peptide Sequencing with Conditional Mutual Information, page 8: Models’ Performance on mass spectrum dataset with syn- datasets are derived from different species, there exists a thetic noise. The results are for the Clam bacteria test set, which is significant difference in the distribution of PTMs quantities.

Nine-species benchmark Amino acid precision Peptide precision
Casanovo · retrained 0.582 0.297
AdaNovo w/oPSM-levelMI 0.586 0.311
AdaNovo w/oAA-levelCMI 0.614 0.335
AdaNovo 0.621 0.342

Table5

AdaNovo: Adaptive De Novo Peptide Sequencing with Conditional Mutual Information, page 8: Comparisons with alternative methods in terms of identi- to reduced precision in identification. To address these isfying amino acids with PTMs. All results are for the yeast test set, sues, we introduce a novel approach involving the calculawhich is one of 9-species benchmark (Tran et al., 2017). tion of conditional mutual information between the spec-

Nine-species benchmark Amino acid precision Peptide precision
Casanovo · retrained 0.753 0.568
Casanovo re-weight · retrained 0.762 0.576
Casanovo focal loss · retrained 0.745 0.543
AdaNovo w/oPSM-levelMI 0.784 0.582
AdaNovo 0.793 0.593

Papers describing it (2)

Authors (12)

Jun Xia, Shaorong Chen, Jingbo Zhou, Tianze Ling, Wenjie Du, Sizhe Liu, Stan Z. Li, Xiaojun Shan, Zhangyang Gao, Cheng Tan, Bozhen Hu, Jiangbin Zheng

Seen in the charts

Back to the full map

Back to top