CNN + RNN
16 methods · 2017–2025
CNN + RNN: The first deep-learning generation: a convolutional network reads the spectrum and a recurrent decoder emits the sequence. This is what made a spectrum something to decode end to end rather than something to search.
The first deep-learning generation: a convolutional network reads the spectrum and a recurrent decoder emits the sequence. This is what made a spectrum something to decode end to end rather than something to search.
The earliest of its 16 methods is DeepNovo (2017); 15 more have followed.
| Methods | 16 |
| Papers describing them | 19 |
| Authors | 85 |
| Active | 2017-07-18 to 2025-02-07 |
| Deep learning | 16 of 16 |
| Kinds | algorithm (11), downstream-application (3), adjacent, post-processor |
| Acquisition | DDA (14), DIA (2) |
Methods (16)
Oldest first, by the paper that describes each one.
- DeepNovo (2017): First DL model (CNN+LSTM)
- Deep-learning protein identification primer (2017): Companion vision paper by the DeepNovo authors laying out how deep learning applies to protein identification end-to-end (from spectrum → amino-acid tokens → peptide → protein). Predates the PNAS DeepNovo paper by a few months.
- DeepNovo-DIA (2018): First de novo for DIA
- DeepNovo V2 (2019): Improved CNN+LSTM model
- DeepNovoAA (2019): Immunopeptidome neoantigen discovery
- Prosit (2019): Deep-learning (bidirectional RNN + attention) predictor of peptide fragment-ion intensities and retention times: widely used to validate and rescore de novo sequencing candidates against predicted spectra.
- SMSNet (2019): Sequence-Mask-Search seq2seq
- PointNovo (2020): Point-set network
- Deep Novo A+ (2022): Improved DeepNovo model
- DePS (2022): Improved deep learning model
- Kaiko (2022): Deep-learning (CNN + RNN) de novo peptide sequencer trained on 5 M peptide-spectrum matches from 55 phylogenetically diverse bacteria. Identifies microbial community members directly from metaproteomic MS/MS, then builds sample-specific protein databases without requiring matched metagenomes: validated on native soil microbiome samples.
- AL-DeepNovo-DIA (2022): Active-learning wrapper around Tran et al. 2019’s DeepNovo-DIA. Uses acquisition functions to pick the most informative spectra at each training iteration instead of random selection. First published as AL-DeepNovo-DIA (ICML 2022 Workshop on Computational Biology, Ebrahimi & Guo); rebranded ActiveNovo_DIA in Ebrahimi’s 2025 UNT PhD thesis.
- NovoRank (2022): Spectral clustering refinement
- Multienzyme DeepNovo models (2023): DeepNovo-based multienzyme models for improving peptide de novo sequencing generalizability.
- PGPointNovo (2023): Parallel GPU-based PointNovo
- DeepNovo Peptidome (2023): Immunopeptidomics pipeline
How they score
3 of the 16 have been run on denovo_benchmarks, which ranks 17 tools over 84 datasets. The family’s best median rank is 12.
- SMSNet: median peptide-level average precision 0.585, median rank 12 of 17
- DePS: median peptide-level average precision 0.458, median rank 15 of 17
- DeepNovo: median peptide-level average precision 0.204, median rank 17 of 17
Read these next to the rest of the field, not on their own: what the numbers mean.
Applied in
Papers describing them (19)
- De novo peptide sequencing by deep learning (2017, PNAS, peer-reviewed)
- Protein identification with deep learning: from abc to xyz (2017, arXiv, preprint)
- Deep learning enables de novo peptide sequencing from data-independent-acquisition mass spectrometry (2018, Nature Methods, peer-reviewed)
- DeepNovoV2: Better de novo peptide sequencing with deep learning (2019, arXiv, preprint)
- Personalized deep learning of individual immunopeptidomes to identify neoantigens for cancer vaccines (2019, bioRxiv, preprint)
- Prosit: proteome-wide prediction of peptide tandem mass spectra by deep learning (2019, Nature Methods, peer-reviewed)
- Uncovering Thousands of New Peptides with Sequence-Mask-Search Hybrid De Novo Peptide Sequencing Framework (2019, Molecular & Cellular Proteomics, peer-reviewed)
- Peptide Sequencing with Deep Learning (2020, thesis)
- Personalized deep learning of individual immunopeptidomes to identify neoantigens for cancer vaccines (2020, Nature Machine Intelligence, peer-reviewed)
- Computationally instrument-resolution-independent de novo peptide sequencing for high-resolution devices (2021, Nature Machine Intelligence, peer-reviewed)
- Deep Novo A+: Improving the Deep Learning Model for De Novo Peptide Sequencing with Additional Ion Types and Validation Set (2022, Current Bioinformatics, peer-reviewed)
- DePS: An improved deep learning model for de novo peptide sequencing (2022, arXiv, preprint)
- Uncovering Hidden Members and Functions of the Soil Microbiome Using De Novo Metaproteomics (2022, Journal of Proteome Research, peer-reviewed)
- Deep Active Learning for De Novo Peptide Sequencing from Data-independent-acquisition Mass Spectrometry (2022, ICML 2022 Workshop on Computational Biology, ML conference)
- NovoRank: Machine Learning Based Post-processing for Performance Improvement in De Novo Peptide Sequencing (2022, thesis)
- Multienzyme deep learning models improve peptide de novo sequencing by mass spectrometry proteomics (2023, PLOS Computational Biology, peer-reviewed)
- PGPointNovo: an efficient neural network-based tool for parallel de novo peptide sequencing (2023, Bioinformatics Advances, peer-reviewed)
- A complete mass spectrometry-based immunopeptidomics pipeline for neoantigen identification and validation (2023, Research Square, preprint)
- NovoRank: Refinement for De Novo Peptide Sequencing Based on Spectral Clustering and Deep Learning (2025, Journal of Proteome Research, peer-reviewed)