ScopeNovo: a de novo peptide sequencing method using spectrum-spectrum similarity to refine the prediction accuracy.
- 1Department of Computer Science, City University of Hong Kong, Hong Kong, China.
Abstract
Motivation
De novo peptide sequencing from tandem mass spectra is essential for characterizing unknown proteins, identifying novel post-translational modifications, and analysing samples without reference databases. While recent deep learning models have improved the predicted peptide accuracy, they typically select the predicted peptide sequence based solely on sequence-level decoding scores. Very often, high scores are assigned to peptides without the direct support from the observed mass spectrum.Results
We present "ScopeNovo", a de novo peptide sequencing method that improves predicted peptide accuracy by using spectrum-spectrum similarity between the input spectrum and the "representative" spectrum corresponding to the predicted peptide sequence. ScopeNovo generates multiple candidate peptides using an existing peptide de novo sequencing learning model and reranks candidates based on spectrum-spectrum similarity. To test our method, we use π-PrimeNovo as the base learning model. Experimental results on the standard nine-species benchmark datasets V1 and V2 show that ScopeNovo can substantially improve the prediction accuracy. We designed two settings, where Setting 1 is for well-annotated species, and Setting 2 is for species that are not well annotated. For Setting 1, ScopeNovo improves the average peptide accuracies of π-PrimeNovo over the nine species by 6.1% and 4.0% on V1 and V2, respectively. For Setting 2, ScopeNovo can still improve the peptide accuracy of π-PrimeNovo over the nine species by 3.4% and 1.9% on the V1 and V2 benchmark datasets, respectively. Compared to the best-known method, RankNovo, the peptide accuracies over the nine species on V1 and V2 of our method for Setting 1 are 0.711 and 0.737, where the peptide accuracies for RankNovo are 0.687 and 0.682. Even for Setting 2, the peptide accuracies of our method are comparable to those of RankNovo.Availability and implementation
ScopeNovo is implemented in Python under an MIT license. The source code is freely available at https://github.com/SoSongzhi/ScopeNovo.git. The test data are available at https://doi.org/10.5281/zenodo.23181841.
Related
- Defining Applicability Domains for Metal–Ligand Selectivity Prediction: Predictive Uncertainty vs. Molecular Similarity
- Ranking versus Absolute Accuracy in Matter–Light Simulation Cost Prediction
- Mechanism-grounded fragmentation trees enable explicit chemical reasoning for electron ionization mass spectrum prediction
- Integrating classical similarity and multitask neural representations
- Drug Molecular and Route Similarity-Aware Retrieval-Augmented LLM Generation with Knowledge Graph for Semantic Reasoning
BibTeX
@article{song2026scopenovo,
title = {{ScopeNovo: a de novo peptide sequencing method using spectrum-spectrum similarity to refine the prediction accuracy.}},
author = {Zhi Song and Zejian Ding and Lusheng Wang},
journal = {Bioinformatics (Oxford, England)},
year = {2026},
month = oct,
doi = {10.1093/bioinformatics/btag755},
url = {https://europepmc.org/article/MED/42879704},
note = {CC BY},
}
RIS
TY - JOUR
TI - ScopeNovo: a de novo peptide sequencing method using spectrum-spectrum similarity to refine the prediction accuracy.
AU - Zhi Song
AU - Zejian Ding
AU - Lusheng Wang
PY - 2026
DA - 2026/10/01
JO - Bioinformatics (Oxford, England)
DO - 10.1093/bioinformatics/btag755
UR - https://europepmc.org/article/MED/42879704
AB - Motivation De novo peptide sequencing from tandem mass spectra is essential for characterizing unknown proteins, identifying novel post-translational modifications, and analysing samples without reference databases. While recent deep learning models have improved the predicted peptide accuracy, they typically select the predicted peptide sequence based solely on sequence-level decoding scores. Very often, high scores are assigned to peptides without the direct support from the observed mass spectrum.Results We present "ScopeNovo", a de novo peptide sequencing method that improves predicted peptide accuracy by using spectrum-spectrum similarity between the input spectrum and the "representative" spectrum corresponding to the predicted peptide sequence. ScopeNovo generates multiple candidate peptides using an existing peptide de novo sequencing learning model and reranks candidates based on spectrum-spectrum similarity. To test our method, we use π-PrimeNovo as the base learning model. Experimental results on the standard nine-species benchmark datasets V1 and V2 show that ScopeNovo can substantially improve the prediction accuracy. We designed two settings, where Setting 1 is for well-annotated species, and Setting 2 is for species that are not well annotated. For Setting 1, ScopeNovo improves the average peptide accuracies of π-PrimeNovo over the nine species by 6.1% and 4.0% on V1 and V2, respectively. For Setting 2, ScopeNovo can still improve the peptide accuracy of π-PrimeNovo over the nine species by 3.4% and 1.9% on the V1 and V2 benchmark datasets, respectively. Compared to the best-known method, RankNovo, the peptide accuracies over the nine species on V1 and V2 of our method for Setting 1 are 0.711 and 0.737, where the peptide accuracies for RankNovo are 0.687 and 0.682. Even for Setting 2, the peptide accuracies of our method are comparable to those of RankNovo.Availability and implementation ScopeNovo is implemented in Python under an MIT license. The source code is freely available at https://github.com/SoSongzhi/ScopeNovo.git. The test data are available at https://doi.org/10.5281/zenodo.23181841.
SN - 1367-4803
SN - 1367-4811
ER -