Pairwise

algorithm · Transformer (AR)

Pairwise: algorithm · Transformer (AR). Mass difference attention

Mass difference attention

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 0.808 over 86 datasets, median rank 4 of 14 (version bm-1.0.0).

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

Reported comparisons (3)

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.

Table2 (Nine-species · original (DeepNovo, 2017))

Pairwise Attention: Leveraging Mass Differences to Enhance De Novo Sequencing of Mass Spectra, page 7: Peptide precision at 100% coverage as measured on the V1 and V2 of the nine-species dataset. Here, Base denotes our implementation of a standard transformer, i.e. without pairwise features. PA is our implementation of pairwise attention, and Casanovo is the reported numbers for Casanovo bm in the original publication7. For PA we report the average of 3 runs at seeds 0, 10, and 20, with standard deviation in parentheses.

Nine-species benchmark
original (DeepNovo, 2017) 		Peptide precision
Apis mellifera Bacillus subtilis Candidatus Thiodiazotropha endoloripes Homo sapiens Methanosarcina mazei Mus musculus Solanum lycopersicum Saccharomyces cerevisiae Vigna mungo Average
Pairwise base 0.390 0.536 0.357 0.340 0.503 0.433 0.509 0.537 0.570 0.464
Pairwise 0.463 0.612 0.409 0.391 0.554 0.472 0.590 0.612 0.625 0.523
Casanovo · quoted 0.433 0.573 0.390 0.383 0.515 0.431 0.522 0.580 0.552 0.487

Table2 (Nine-species · revised (main))

Pairwise Attention: Leveraging Mass Differences to Enhance De Novo Sequencing of Mass Spectra, page 7: Peptide precision at 100% coverage as measured on the V1 and V2 of the nine-species dataset. Here, Base denotes our implementation of a standard transformer, i.e. without pairwise features. PA is our implementation of pairwise attention, and Casanovo is the reported numbers for Casanovo bm in the original publication7. For PA we report the average of 3 runs at seeds 0, 10, and 20, with standard deviation in parentheses.

Nine-species benchmark
revised (main) 		Peptide precision
Apis mellifera Bacillus subtilis Candidatus Thiodiazotropha endoloripes Homo sapiens Methanosarcina mazei Mus musculus Solanum lycopersicum Saccharomyces cerevisiae Vigna mungo Average
Pairwise base 0.390 0.494 0.356 0.451 0.509 0.410 0.543 0.558 0.525 0.471
Pairwise 0.446 0.583 0.425 0.521 0.579 0.435 0.623 0.631 0.598 0.538
Casanovo · quoted 0.456 0.538 0.468 0.533 0.529 0.395 0.608 0.561 0.428 0.502

Table3

Pairwise Attention: Leveraging Mass Differences to Enhance De Novo Sequencing of Mass Spectra, page 8: Performance after training the models on the MassIVE-KB set. Here we tested on each of the species in the nine-species dataset and report peptide precision at 100% coverage. The models tested are Base and PA, alongside Casanovo’s reported numbers7.

Nine-species benchmark
revised (main) 		Peptide precision at coverage 1
Apis mellifera Bacillus subtilis Candidatus Thiodiazotropha endoloripes Homo sapiens Methanosarcina mazei Mus musculus Solanum lycopersicum Saccharomyces cerevisiae Vigna mungo Average
Pairwise base 0.618 0.706 0.525 0.737 0.700 0.563 0.735 0.765 0.753 0.678
Pairwise 0.640 0.732 0.549 0.746 0.720 0.579 0.751 0.784 0.783 0.698
Casanovo · quoted 0.662 0.778 0.656 0.740 0.710 0.552 0.799 0.840 0.762 0.722

Papers describing it (2)

Authors (4)

Joel Lapin, Alfred Nilsson, Mathias Wilhelm, Lukas Käll

Seen in the charts

Back to the full map

Back to top