Pith. sign in

REVIEW 3 major objections 4 minor 47 references

Latent Imputation before Prediction: A New Computational Paradigm for De Novo Peptide Sequencing

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read LIPNovo demonstrates that imputing the latent representations of missing b/y fragment peaks before peptide prediction improves de novo sequencing across three benchmarks, reaching 0.797 amino acid precision on the Nine-species dataset.

desk verdict LIPNovo gets real gains on de novo sequencing benchmarks, but the paper hasn't shown the imputation content is what causes them — the ablations tangle the imputation module with extra decoder tokens. read the letter →

arxiv 2505.17524 v1 pith:L3F6IDSA submitted 2025-05-23 cs.CE

Ye Du , Chen Yang , Nanxi Yu , Wanyu Lin , Qian Zhao , Shujun Wang This is my paper
classification cs.CE MSC 68T0792C40
keywords denovopeptidesequencingtandemmassspectrometrymissingfragmentationimputationlatentrepresentationsetpredictionwithbipartitematchingtransformerdecoderproteomicsLIPNovo
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that a major bottleneck in de novo peptide sequencing is missing fragmentation information: real MS/MS spectra often lack peaks for b/y ions that should be present. It proposes LIPNovo, a pipeline that inserts an explicit imputation step between spectrum encoding and peptide decoding, predicting the missing peaks in latent space rather than as raw m/z values. During training, the imputation targets come from a theoretical spectrum computed from the ground-truth peptide, so no extra annotation is needed. On three benchmark datasets this raises amino acid-level precision from 0.741 to 0.797 (Nine-species), 0.357 to 0.557 (Seven-species), and 0.525 to 0.637 (HC-PT) relative to the retrained baseline, and it outperforms published state-of-the-art methods on amino acid, peptide, and PTM metrics. The authors attribute the gain to the imputation mechanism itself rather than to the added parameters.

What carries the argument

The central object is the latent imputation module $\Phi_\theta$: a transformer decoder with $M=100$ learnable peak queries that takes the observed spectrum representations $z$ as encoded memory and outputs candidate latent vectors plus confidence scores. Training uses a theoretical spectrum $z'$ built from the ground-truth peptide, assuming every b- and y-ion has charge $+1$ and constant intensity equal to the maximum observed intensity; optimal bipartite matching assigns each predicted vector to a theoretical peak or to $\varnothing$, and the imputation loss combines MSE on matched pairs with a cross-entropy term over assignments. This machinery makes missing-fragment recovery an explicit, end-to-end trainable objective before peptide decoding.

What would settle it

Train LIPNovo on spectra whose fragmentation is dominated by multiply charged ions or c/z ions, such as ETD data, and compare against the same baseline: if the imputation stage stops providing gains, or if the imputed latent vectors no longer match actual missing fragment identities when scored against true peak labels, the central claim is refuted. A cheaper check on existing data is to replace the constant-intensity, +1-charge theoretical spectrum with an empirically calibrated fragmentation model and see whether imputation quality and peptide accuracy improve or degrade.

Watch

Extended reading notes

Core claim

LIPNovo's central claim is that compensating for missing fragmentation before prediction, in latent space, is a better computational paradigm than asking an encoder-decoder model to work directly with incomplete spectra. The paper frames imputation as a set-prediction problem: a transformer decoder with a fixed set of learnable peak queries reads the observed spectrum representation and outputs candidate latent vectors, which are matched to the latent representations of theoretical b/y ions by optimal bipartite matching using the Hungarian algorithm. The imputation loss combines mean squared error on matched latent vectors with a classification term for predictions assigned to the empty set. During inference, the theoretical spectrum is discarded; the filtered imputed representations are concatenated with the observed spectrum representation and fed to the autoregressive peptide decoder. The result is consistent gains across all three benchmarks, with ablations showing that the imputation module contributes most of the improvement.

Load-bearing premise

The load-bearing assumption is that an idealized theoretical spectrum—every b/y ion present, singly charged, with a single uniform intensity—is a good training target for what real fragmentation should look like; if real instruments deviate systematically from that ideal, the imputation module is trained to fill in the wrong picture and the gains may not transfer.

Editorial extensions

If this is right

  • Adding an explicit latent imputation stage before decoding becomes a viable design for de novo sequencing; the reported gains are +5.6%, +20.0%, and +11.2% amino acid precision over the retrained baseline on Nine-species, Seven-species, and HC-PT.
  • Because the upper bound from using ground-truth theoretical representations is 0.939 amino acid precision, continued improvement of imputation quality is a concrete path to further sequencing gains.
  • The imputation targets are derived from the same ground-truth peptides used for the sequence loss, so the paradigm can be dropped into existing encoder-decoder training pipelines without new annotations.
  • The ablation and parameter-matched comparison imply that the gain comes from the imputation objective itself, not from added parameters, so comparable gains should require an explicit imputation mechanism rather than simply larger models.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The same latent-imputation-before-prediction idea could extend to other sequencing tasks where a theoretical fragmentation target can be enumerated, such as oligonucleotide or glycan mass spectrometry.
  • The idealized +1 b/y target is the main transfer risk; adapting the theoretical spectrum to instrument-specific fragmentation, for example including c/z ions or neutral losses, is a natural testable extension that would show whether the gains survive outside HCD-style data.
  • High missing-fragmentation ratios above 0.8 show sharply reduced gains in the paper's own analysis, suggesting that this paradigm will need external priors or cross-sample information in the most degraded spectra.
  • One could probe whether the imputed latents genuinely encode fragment identity by decoding them into peak-level m/z predictions or classifying them against true fragment types, a check the paper does not perform.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. LIPNovo introduces a latent-space imputation step for de novo peptide sequencing: after encoding the observed MS/MS spectrum, a transformer-decoder imputation module with learned peak queries predicts latent representations of an idealized theoretical b/y-ion spectrum; filtered imputed vectors are concatenated with the observed spectrum tokens and passed to an autoregressive peptide decoder. Training uses bipartite matching plus an imputation loss, and also supervises the decoder on the theoretical-spectrum context. Experiments on the Nine-species, Seven-species, and HC-PT datasets report state-of-the-art amino acid-, peptide-, and PTM-level results, together with leave-one-out cross-validation, hyperparameter sensitivity, missing-fragmentation-ratio analyses, and a parameter-matched baseline.

Significance. If the performance gains are attributable to the content of the imputed theoretical-peak representations, this is a meaningful advance: it offers a simple plug-in module for existing encoder-decoder architectures, with released code and consistent gains across three benchmark datasets. The paper also provides useful auxiliary evidence, including leave-one-out species generalization, a hyperparameter sensitivity study, and an oracle upper-bound analysis. However, the causal attribution to theoretical-peak imputation is not yet established, because none of the ablation or oracle conditions controls for the increased number of tokens presented to the peptide decoder. The central mechanistic claim therefore needs additional control experiments before the paper can support its strongest conclusion.

major comments (3)
  1. [Section 4.4, Table 5; Section 3.2.5, Eq. (7)] The ablations do not control for the number of tokens fed to the peptide decoder. Row 2 of Table 5 (baseline + complementary spectrum) passes only observed tokens, while rows 3-5 additionally concatenate up to M=100 imputation-module outputs filtered by the threshold in Eq. (7). The improvement from row 2 to row 3 (+1.1% amino acid precision) and from row 4 to row 5 (+1.5%) is therefore compatible with a model that benefits simply from additional context vectors, regardless of whether those vectors encode theoretical peaks. The oracle experiment in Figure 5 has the same confound, because substituting ground-truth theoretical representations also increases the token count. Please add control conditions that feed the same number of non-informative tokens (for example, random latent vectors, duplicated observed peaks, or additional complementary-spectrum embeddings) to the decoder; only if the imputed vectors content matters beyond token count can the paper claim that latent imputation drives the gain.
  2. [Section 4.4, Table 5] The claim that 'using the imputation mechanism alone leads to significant improvements' is based on row 4 versus row 1, but row 4 simultaneously adds the imputation module and the auxiliary cross-entropy loss LCE(z') from Eq. (9). Row 4 therefore does not isolate the imputation mechanism. The comparison that removes LCE(z') (row 3 versus row 2) shows only a +1.1% amino acid precision gain, and even that comparison is still subject to the token-count confound described above. Please report an ablation that adds LCE(z') to the baseline without imputation tokens, or otherwise orthogonalizes the auxiliary loss from the token-count increase.
  3. [Section 3.2.2, Eq. (3); Section 3.2.3, Eq. (4)] The imputation targets are generated under the assumptions that every theoretical fragment is singly charged and that its intensity equals the maximum observed intensity (I'_j = max{I_1, ..., I_N}). Real tandem mass spectra contain multiply charged ions, neutral losses, and internal ions, so this is an idealized target that may deviate systematically from the instrument's actual fragmentation behavior. The paper does not quantify how often singly charged b/y ions dominate in the three datasets, nor does it test transfer to data with different collision energies or instrument types. Please add a diagnostic (for example, the proportion of observed peaks attributable to singly charged b/y ions, or an evaluation on a dataset with substantial multi-charge fragmentation) to justify the target's validity for the claimed general paradigm.
minor comments (4)
  1. [Throughout] There are several typos and wording issues: 'w.r.tthe' in Section 1, 'tailed imputation training objective' in Figure 2 and Equation (9) should be 'tailored', and 'the fist M peaks' in Section 4.4 should be 'the first M peaks'.
  2. [Equation (5)] The indicator notation in Equation (5) is malformed as typeset; please use standard notation such as 1{z'_j not in empty-set} and 1{z'_j in empty-set} for clarity.
  3. [Figure 3] The peptide label in Figure 3 is hard to read (for example, 'PETIDP' in the inset); please enlarge and align the fragment labels with the sequence positions so that the b-ion and y-ion derivation is unambiguous.
  4. [Tables 1-3, 5, 6] All reported numbers are from single runs without variance estimates. Given that some improvements are small (for example, Table 6 shows differences of around 0.005, and Table 3 shows a PTM recall decrease on some species), please provide repeated-seed means and standard deviations for at least the main comparisons and core ablations.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: imputation targets are training-only and test-time predictions use only observed spectra.

full rationale

The central derivation is not circular. The imputation targets z' are constructed from the ground-truth peptide sequence (Section 3.2.2, Eq. 3) and are used only during training; at inference 'the upper part is discarded, eliminating the need for the theoretical spectrum during testing' (Figure 2 caption and Remark 3.1). The peptide decoder is trained with LCE([z~; z]) + LCE(z') + L_Imputation (Eq. 9), but at test time the prediction uses only the observed-spectrum representations z and the imputed representations z~ produced by Phi_theta, with no ground-truth label reused. The upper-bound experiment in Figure 5 is explicitly labeled as an upper bound obtained by directly using ground-truth representations, so it is an oracle diagnostic rather than a claimed prediction. No fitted parameter is renamed as a prediction, no load-bearing argument rests on a self-citation, and no uniqueness theorem is imported from the authors' prior work. The ablations in Table 5 do alter decoder token count when the imputation module is added, which is a potential experimental confound, but that is a validity concern rather than a circularity: the reported gains are measured on held-out test spectra and are not forced by construction.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the validity of the constructed theoretical spectrum as a training target and on the learnability of missing fragmentation patterns. No new physical entities are introduced. Four model hyperparameters and one intensity modeling choice are selected by hand or validation, but they are standard architecture choices rather than fitted physical constants.

free parameters (4)
  • number of peak queries M = 100
    Tuned on the Nine-species validation set; optimal at M=100 in Table 6, used consistently across datasets.
  • imputation filter threshold tau = 0.8
    Tuned on the Nine-species validation set; optimal at tau=0.8 in Table 6.
  • imputation decoder layer count = 3
    Tuned on the Nine-species validation set; optimal at 3 layers in Table 6.
  • theoretical peak intensity I'_j = max observed intensity
    Hand-set modeling choice in Section 3.2.2, based on the prior that ideal fragmentation should occur at consistently high intensity.
assumptions (5)
  • domain assumption Peptide fragmentation primarily produces b- and y-ions at charge +1.
    The theoretical spectrum used as the imputation target consists only of b/y ions and assumes charge +1 for every fragment (Section 3.2.2).
  • domain assumption A theoretical spectrum computed from the ground-truth peptide is a reasonable supervision target for imputation.
    The imputation loss minimizes distance to latent representations of this constructed spectrum, and LCE(z') forces those representations to be decodable to the peptide.
  • domain assumption The same spectrum encoder can represent both observed and theoretical spectra in a shared latent space.
    Weight sharing between observed and theoretical encoding is shown in Figure 2, so z and z' must be comparable for the MSE loss to be meaningful.
  • domain assumption Missing fragmentation patterns are learnable across the training distribution.
    Remark 3.1 explicitly states that the model learns missing patterns from large-scale data and cross-instance correlations.
  • standard math Standard transformer self-attention and cross-attention are sufficient to model peak interactions.
    The architecture follows CasaNovo and relies on the established expressivity of transformer attention for spectra.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Latent Imputation before Prediction: A New Computational Paradigm for De Novo Peptide Sequencing." pith.science (2026). https://pith.science/paper/L3F6IDSA

@misc{pith2026250517524,
  author       = {Pith},
  title        = {Pith review of: Latent Imputation before Prediction: A New Computational Paradigm for De Novo Peptide Sequencing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/L3F6IDSA}},
  note         = {Machine review of arXiv:2505.17524}
}
read the original abstract

De novo peptide sequencing is a fundamental computational technique for ascertaining amino acid sequences of peptides directly from tandem mass spectrometry data, eliminating the need for reference databases. Cutting-edge models usually encode the observed mass spectra into latent representations from which peptides are predicted autoregressively. However, the issue of missing fragmentation, attributable to factors such as suboptimal fragmentation efficiency and instrumental constraints, presents a formidable challenge in practical applications. To tackle this obstacle, we propose a novel computational paradigm called \underline{\textbf{L}}atent \underline{\textbf{I}}mputation before \underline{\textbf{P}}rediction (LIPNovo). LIPNovo is devised to compensate for missing fragmentation information within observed spectra before executing the final peptide prediction. Rather than generating raw missing data, LIPNovo performs imputation in the latent space, guided by the theoretical peak profile of the target peptide sequence. The imputation process is conceptualized as a set-prediction problem, utilizing a set of learnable peak queries to reason about the relationships among observed peaks and directly generate the latent representations of theoretical peaks through optimal bipartite matching. In this way, LIPNovo manages to supplement missing information during inference and thus boosts performance. Despite its simplicity, experiments on three benchmark datasets demonstrate that LIPNovo outperforms state-of-the-art methods by large margins. Code is available at \href{https://github.com/usr922/LIPNovo}{https://github.com/usr922/LIPNovo}.

Figures

Figures reproduced from arXiv: 2505.17524 by the authors.

Figure 1
Figure 1. Comparison of amino acid-level precision between LIP￾Novo (ours) and existing methods under varying missing frag￾mentation ratios. As the missing ratio increases, performance deteriorates dramatically, highlighting the detrimental impact of the missing fragmentation issue. The proposed LIPNovo con￾sistently outperforms existing methods across all missing ratios. Results are based on the test set (i.e., the yeast spe… view at source ↗
Figure 2
Figure 2. Illustration of the computational paradigm of LIPNovo. During training, LIPNovo generates a theoretical spectrum based on the target peptide ( [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustration of theoretical spectrum calculation. For ex￾ample, by splitting the position at ‘E’ and ‘P’, we can derive the b2 ion (PE) and the y5 ion (PTIDE). The masses of these two ions can be calculated using the mass table of amino acid residues. Here, we assume a charge of +1 and set the intensity to 100%. As such, we can get a set of mass spectrum representation, denoted as z = {zi} N i=1, where each zi ∈ R d… view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Imputation quality vs. model performance. (Left) A smaller imputation loss corresponds to higher performance. (Right) The upper bound of LIPNovo, obtained by directly using ground truth representations instead of predicted representations for the theoretical spectrum o…
Figure 6
Figure 6. Figure 6: Illustration of the identification workflow of shotgun proteomics (Wolters et al., 2001). B. More Details [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]
Figure 7
Figure 7. Figure 7: Comparison of amino acid recall (left column) and peptide recall (right column) under different missing fragmentation ratios between LIPNovo and state-of-the-art methods on three datasets. 13 [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 30 canonical work pages

  1. [1]

    and Mann, M

    Aebersold, R. and Mann, M. Mass spectrometry-based proteomics. Nature, 422 0 (6928): 0 198--207, 2003

    work page 2003

  2. [2]

    Bandeira, N., Pham, V., Pevzner, P., Arnott, D., and Lill, J. R. Automated de novo protein sequencing of monoclonal antibodies. Nature biotechnology, 26 0 (12): 0 1336--1338, 2008

    work page 2008

  3. [3]

    Blackstock, W. P. and Weir, M. P. Proteomics: quantitative and physical mapping of cellular proteins. Trends in biotechnology, 17 0 (3): 0 121--127, 1999

    work page 1999

  4. [4]

    End-to-end object detection with transformers

    Carion, N., Massa, F., Synnaeve, G., Usunier, N., Kirillov, A., and Zagoruyko, S. End-to-end object detection with transformers. In European conference on computer vision, pp.\ 213--229. Springer, 2020

    2020

  5. [5]

    B., Williams, W., van Beljouw, S

    Eloff, K., Kalogeropoulos, K., Morell, O., Mabona, A., Jespersen, J. B., Williams, W., van Beljouw, S. P., Skwark, M., Laustsen, A. H., Brouns, S. J., et al. De novo peptide sequencing with instanovo: Accurate, database-free peptide identification for large scale proteomics experiments. bioRxiv, pp.\ 2023--08, 2023

    work page 2023

  6. [6]

    and Al-Onaizan, Y

    Freitag, M. and Al-Onaizan, Y. Beam search strategies for neural machine translation. arXiv preprint arXiv:1702.01806, 2017

    arXiv 2017

  7. [7]

    Generative adversarial networks

    Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. Generative adversarial networks. Communications of the ACM, 63 0 (11): 0 139--144, 2020

    2020

  8. [8]

    and Hasan, M

    Hayat, A. and Hasan, M. R. Claim your data: Enhancing imputation accuracy with contextual large language models. arXiv preprint arXiv:2405.17712, 2024

    arXiv 2024

Show all 47 references
  1. [9]

    Contranovo: A contrastive learning approach to enhance de novo peptide sequencing

    Jin, Z., Xu, S., Zhang, X., Ling, T., Dong, N., Ouyang, W., Gao, Z., Chang, C., and Sun, S. Contranovo: A contrastive learning approach to enhance de novo peptide sequencing. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pp.\ 144--152, 2024

    2024

  2. [10]

    Kingma, D. P. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013

    2013 arXiv

  3. [11]

    Kuhn, H. W. The hungarian method for the assignment problem. Naval research logistics quarterly, 2 0 (1-2): 0 83--97, 1955

    1955

  4. [12]

    Deep learning

    LeCun, Y., Bengio, Y., and Hinton, G. Deep learning. nature, 521 0 (7553): 0 436--444, 2015

    2015

  5. [13]

    Predicting protein function from sequence and structure

    Lee, D., Redfern, O., and Orengo, C. Predicting protein function from sequence and structure. Nature reviews molecular cell biology, 8 0 (12): 0 995--1005, 2007

    2007

  6. [14]

    Comparison of the effects of imputation methods for missing data in predictive modelling of cohort study datasets

    Li, J., Guo, S., Ma, R., He, J., Zhang, X., Rui, D., Ding, Y., Li, Y., Jian, L., Cheng, J., et al. Comparison of the effects of imputation methods for missing data in predictive modelling of cohort study datasets. BMC Medical Research Methodology, 24 0 (1): 0 41, 2024 a

    2024

  7. [15]

    Method for incomplete and imbalanced data based on multivariate imputation by chained equations and ensemble learning

    Li, J., Wang, Z., Wu, L., Qiu, S., Zhao, H., Lin, F., and Zhang, K. Method for incomplete and imbalanced data based on multivariate imputation by chained equations and ensemble learning. IEEE Journal of Biomedical and Health Informatics, 2024 b

    2024

  8. [16]

    Relevant applications of generative adversarial networks in drug design and discovery: molecular de novo design, dimensionality reduction, and de novo peptide and protein design

    Lin, E., Lin, C.-H., and Lane, H.-Y. Relevant applications of generative adversarial networks in drug design and discovery: molecular de novo design, dimensionality reduction, and de novo peptide and protein design. Molecules, 25 0 (14): 0 3250, 2020

    2020

  9. [17]

    and Tsai, C.-F

    Lin, W.-C. and Tsai, C.-F. Missing value imputation: a review and analysis of the literature (2006--2017). Artificial Intelligence Review, 53: 0 1487--1509, 2020

    2006

  10. [18]

    Accurate de novo peptide sequencing using fully convolutional neural networks

    Liu, K., Ye, Y., Li, S., and Tang, H. Accurate de novo peptide sequencing using fully convolutional neural networks. Nature Communications, 14 0 (1): 0 7974, 2023

    2023

  11. [19]

    Peaks: powerful software for peptide de novo sequencing by tandem mass spectrometry

    Ma, B., Zhang, K., Hendrie, C., Liang, C., Li, M., Doherty-Kirby, A., and Lajoie, G. Peaks: powerful software for peptide de novo sequencing by tandem mass spectrometry. Rapid communications in mass spectrometry, 17 0 (20): 0 2337--2342, 2003

    2003

  12. [20]

    Recent advances in mass spectrometry based clinical proteomics: applications to cancer research

    Macklin, A., Khan, S., and Kislinger, T. Recent advances in mass spectrometry based clinical proteomics: applications to cancer research. Clinical proteomics, 17 0 (1): 0 17, 2020

    2020

  13. [21]

    Mitigating the missing-fragmentation problem in de novo peptide sequencing with a two-stage graph-based deep learning model

    Mao, Z., Zhang, R., Xin, L., and Li, M. Mitigating the missing-fragmentation problem in de novo peptide sequencing with a two-stage graph-based deep learning model. Nature Machine Intelligence, 5 0 (11): 0 1250--1260, 2023

    2023

  14. [22]

    McDonald, W. H. and Yates Iii, J. R. Shotgun proteomics and biomarker discovery. Disease markers, 18 0 (2): 0 99--105, 2002

    2002

  15. [23]

    The impact of noise and missing fragmentation cleavages on de novo peptide identification algorithms

    McDonnell, K., Howley, E., and Abram, F. The impact of noise and missing fragmentation cleavages on de novo peptide identification algorithms. Computational and Structural Biotechnology Journal, 20: 0 1402--1412, 2022

    2022

  16. [24]

    I., Vitek, O., and Aebersold, R

    Nesvizhskii, A. I., Vitek, O., and Aebersold, R. Analysis and validation of proteomic data generated by tandem mass spectrometry. Nature methods, 4 0 (10): 0 787--797, 2007

    2007

  17. [25]

    B., Baker, M

    Noor, Z., Ahn, S. B., Baker, M. S., Ranganathan, S., and Mohamedali, A. Mass spectrometry--based protein identification in proteomics—a review. Briefings in bioinformatics, 22 0 (2): 0 1620--1638, 2021

    2021

  18. [26]

    K., Sun, F., White, A

    Nowinski, A. K., Sun, F., White, A. D., Keefe, A. J., and Jiang, S. Sequence, structure, and function of peptide self-assembled monolayers. Journal of the American Chemical Society, 134 0 (13): 0 6000--6005, 2012

    2012

  19. [27]

    Olsen, J. V. and Mann, M. Improved peptide identification in proteomics by two consecutive stages of mass spectrometric fragmentation. Proceedings of the National Academy of Sciences, 101 0 (37): 0 13417--13422, 2004

    2004

  20. [28]

    H., Xin, L., Chen, X., Li, M., Shan, B., and Ghodsi, A

    Qiao, R., Tran, N. H., Xin, L., Chen, X., Li, M., Shan, B., and Ghodsi, A. Computationally instrument-resolution-independent de novo peptide sequencing for high-resolution devices. Nature Machine Intelligence, 3 0 (5): 0 420--425, 2021

    2021

  21. [29]

    Stekhoven, D. J. and B \"u hlmann, P. Missforest—non-parametric missing value imputation for mixed-type data. Bioinformatics, 28 0 (1): 0 112--118, 2012

    2012

  22. [30]

    T., Bludau, I., Zeng, W.-F., Voytik, E., Ammar, C., Schessner, J

    Strauss, M. T., Bludau, I., Zeng, W.-F., Voytik, E., Ammar, C., Schessner, J. P., Ilango, R., Gill, M., Meier, F., Willems, S., et al. Alphapept: a modern and open framework for ms-based proteomics. Nature Communications, 15 0 (1): 0 2168, 2024

    2024

  23. [31]

    Deep learning versus conventional methods for missing data imputation: A review and comparative study

    Sun, Y., Li, J., Xu, Y., Zhang, T., and Wang, X. Deep learning versus conventional methods for missing data imputation: A review and comparative study. Expert Systems with Applications, 227: 0 120201, 2023

    2023

  24. [32]

    L., Smith, L

    Tabb, D. L., Smith, L. L., Breci, L. A., Wysocki, V. H., Lin, D., and Yates, J. R. Statistical characterization of ion trap tandem mass spectra from doubly charged tryptic peptides. Analytical chemistry, 75 0 (5): 0 1155--1163, 2003

    2003

  25. [33]

    H., Zhang, X., Xin, L., Shan, B., and Li, M

    Tran, N. H., Zhang, X., Xin, L., Shan, B., and Li, M. De novo peptide sequencing by deep learning. Proceedings of the National Academy of Sciences, 114 0 (31): 0 8247--8252, 2017

    2017

  26. [34]

    Uzozie, A. C. and Aebersold, R. Advancing translational research and precision medicine with targeted proteomics. Journal of proteomics, 189: 0 1--10, 2018

    2018

  27. [35]

    Attention is all you need

    Vaswani, A. Attention is all you need. Advances in Neural Information Processing Systems, 2017

    2017

  28. [36]

    Recent developments in mass-spectrometry-based targeted proteomics of clinical cancer biomarkers

    Wenk, D., Zuo, C., Kislinger, T., and Sepiashvili, L. Recent developments in mass-spectrometry-based targeted proteomics of clinical cancer biomarkers. Clinical Proteomics, 21 0 (1): 0 6, 2024

    2024

  29. [37]

    R., Royston, P., and Wood, A

    White, I. R., Royston, P., and Wood, A. M. Multiple imputation using chained equations: issues and guidance for practice. Statistics in medicine, 30 0 (4): 0 377--399, 2011

    2011

  30. [38]

    A., Washburn, M

    Wolters, D. A., Washburn, M. P., and Yates, J. R. An automated multidimensional protein identification technology for shotgun proteomics. Analytical chemistry, 73 0 (23): 0 5683--5690, 2001

    2001

  31. [39]

    Xia, J., Chen, S., Zhou, J., Shan, X., Du, W., Gao, Z., Tan, C., Hu, B., Zheng, J., and Li, S. Z. Adanovo: Towards robust emph \ De Novo \ peptide sequencing in proteomics against data biases. In The Thirty-eighth Annual Conference on Neural Information Processing Systems, 2024

    2024

  32. [40]

    Introducing -helixnovo for practical large-scale de novo peptide sequencing

    Yang, T., Ling, T., Sun, B., Liang, Z., Xu, F., Huang, X., Xie, L., He, Y., Li, L., He, F., et al. Introducing -helixnovo for practical large-scale de novo peptide sequencing. Briefings in Bioinformatics, 25 0 (2): 0 bbae021, 2024

    2024

  33. [41]

    R., Eng, J

    Yates, J. R., Eng, J. K., and McCormack, A. L. Mining genomes: correlating tandem mass spectra of modified and unmodified peptides to sequences in nucleotide databases. Analytical chemistry, 67 0 (18): 0 3202--3210, 1995

    1995

  34. [42]

    Yilmaz, M., Fondrie, W., Bittremieux, W., Oh, S., and Noble, W. S. De novo mass spectrometry peptide sequencing with a transformer model. In International Conference on Machine Learning, pp.\ 25514--25522. PMLR, 2022

    2022

  35. [43]

    E., Bittremieux, W., Melendez, C

    Yilmaz, M., Fondrie, W. E., Bittremieux, W., Melendez, C. F., Nelson, R., Ananth, V., Oh, S., and Noble, W. S. Sequence-to-sequence translation from mass spectra to peptides with a transformer model. Nature communications, 15 0 (1): 0 6427, 2024

    2024

  36. [44]

    Nearest neighbor selection for iteratively knn imputation

    Zhang, S. Nearest neighbor selection for iteratively knn imputation. Journal of Systems and Software, 85 0 (11): 0 2541--2552, 2012

    2012

  37. [45]

    and Charoenphakdee, N

    Zheng, S. and Charoenphakdee, N. Diffusion models for missing value imputation in tabular data. arXiv preprint arXiv:2210.17128, 2022

    2022 arXiv

  38. [46]

    Zhou, J., Chen, S., Xia, J., Liu, S., Ling, T., Du, W., Liu, Y., Yin, J., and Li, S. Z. Novobench: Benchmarking deep learning-based emph \ De Novo \ sequencing methods in proteomics. In The Thirty-eight Conference on Neural Information Processing Systems Datasets and Benchmark...

    2024

  39. [47]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.