Evaluation of the accuracy of false alarm frequency control methods for de novo spectrum
peer-reviewed · Proceedings of the Southwest State University. Series: IT Management, Computer Science, Computer Engineering. Medical Equipment Engineering · 2025
| Date | 2025-11-22 |
| Type | peer-reviewed |
| Venue | Proceedings of the Southwest State University. Series: IT Management, Computer Science, Computer Engineering. Medical Equipment Engineering |
| Publisher | Southwest State University |
| Contribution | benchmark |
| DOI | 10.21869/2223-1536-2025-15-3-122-141 |
| Citations (OpenAlex) | 0 |
Abstract
The purpose of the research is comparison of machine learning-based approaches (deep learning) and classical methods for mass spectrum annotation in big data conditions, as well as identification of the optimal scenario for their integration. Methods . The study is based on the PXD004452 dataset containing 2,5 million unique peptides. An interaction scheme based on Python/TensorFlow/PyTorch has been developed, which provides parallel processing of peptide spectra on a GPU cluster. The following steps were used: filtering of the top 150 peaks by intensity; generation of theoretical B-/Y-ions, taking into account modifications; prediction of peptides (PepNet – convolutional+recurrent network; Tidesearch – index-shifting strategy). Metrics: number of matches, delta mass, Levenshtein distance, ROC curves, error distribution. Results . PepNet requires significant computational resources, while the prediction quality is inferior to Tide-search, especially for long peptides and modifications (~average match: 4,2 pi vs 9,7; p 5). Conclusions . The deep learning (PepNet) method shows promise, but without integration with database search, it is inferior in accuracy. A hybrid architecture is proposed: pep-tagging via PepNet, followed by refinement and verification via database search. Such a big data pipeline will combine the discovery of new peptides (de novo) and high identification reliability (database search).
Methods and tools
- PepNet vs Tide false-discovery evaluation: Compares PepNet de novo sequencing with Tide search on PXD004452, finding PepNet less accurate overall but better on spectra whose peptides are absent from the database, and proposes de novo tagging plus database verification.
Methods it uses
- PepNet: Temporal convolutional network
Data used
- HeLa proteome of 12,250 protein-coding genes (as deposited) · PXD004452
Cites (11)
- Latent Imputation before Prediction: A New Computational Paradigm for De Novo Peptide Sequencing (2025) crossref
- π-PrimeNovo: an accurate and efficient non-autoregressive deep learning model for de novo peptide sequencing (2025) both
- PowerNovo: de novo peptide sequencing via tandem mass spectrometry using an ensemble of transformer and BERT models (2024) both
- Transformer-Based De Novo Peptide Sequencing for Data-Independent Acquisition Mass Spectrometry (2023) crossref
- Accurate de novo peptide sequencing using fully convolutional neural networks (2023) both
- Algorithms for de-novo sequencing of peptides by tandem mass spectrometry: A review (2023) both
- DPST: De Novo Peptide Sequencing with Amino-Acid-Aware Transformers (2022) crossref
- DePS: An improved deep learning model for de novo peptide sequencing (2022) crossref
- Complementary Methods for de Novo Monoclonal Antibody Sequencing to Achieve Complete Sequence Coverage (2020) both
- Integrating Database Search and De Novo Sequencing for Immunopeptidomics with DIA Approach (2019) crossref
- De novo peptide sequencing by deep learning (2017) both