Research map / Large language models / Agents, code and math
NovoBench: Benchmarking Deep Learning-based \emph{De Novo} Sequencing Methods in Proteomics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces NovoBench, a unified benchmark that integrates six deep learning de novo peptide sequencing models and evaluates them on three datasets with new PTM, confidence, efficiency, and robustness metrics.
Problem
Deep learning de novo peptide sequencing lacks consensus evaluation datasets, and current metrics (amino acid and peptide precision/recall) ignore PTM identification, efficiency, and robustness to peptide length, noise, and missing fragmentation.
Approach
A unified benchmark that fixes datasets (Seven-species, Nine-species, HC-PT), re-implements six representative models under identical settings, and adds PTM-level precision/recall, a confidence score, AUC, efficiency measures, and factor-robustness analyses (peptide length, noise signal ratio, missing fragmentation ratio).
Key results
- π-HelixNovo achieves the best amino acid-level precision/recall on Nine-species (0.765/0.758) and the best peptide-level precision and AUC on Seven-species (0.234/0.173).
- InstaNovo reaches the highest peptide-level precision on Nine-species (0.532) and HC-PT (0.572) and the highest amino acid-level precision on HC-PT (0.714).
- Peptide-level precision declines significantly as missing fragmentation ratio increases, and InstaNovo is markedly less robust to peptide length than other models.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use π-HelixNovo's complementary spectrum input | π-HelixNovo · Seven-species and Nine-species | peptide-level precision | Helped |
| Increase peptide length | de novo sequencing models on Nine-species · Nine-species | peptide-level precision | Hurt |
| Increase missing fragmentation ratio | de novo sequencing models on Nine-species · Nine-species | peptide-level precision | Hurt |
| Increase noise signal ratio | de novo sequencing models on Nine-species · Nine-species | peptide-level precision | Mixed |
| Use the official pre-trained InstaNovo model rather than training from scratch | InstaNovo · Nine-species and HC-PT | peptide-level precision | Helped |
Surprising findings
- Noise peaks show an initial performance increase with rising NSR, attributed to spectra with low NSR having more missing fragmentations.
- Traditional DeepNovo and PointNovo remain highly competitive with Transformer-based models despite fewer parameters.
Limitations
- Limited to DDA data; DIA data and DIA-specific models are not covered.
- Only three datasets and six models are included.
- InstaNovo was used as an official pre-trained model on Nine-species and HC-PT rather than retrained, complicating comparison.
Experimental setup
Datasets: Seven-species Nine-species HC-PT
Benchmarks: Seven-species Nine-species HC-PT
Metrics: amino acid-level precision amino acid-level recall peptide-level precision AUC PTM precision PTM recall confidence training time inference time trainable parameters
Baselines: DeepNovo PointNovo CasaNovo InstaNovo AdaNovo π-HelixNovo
Builds on: DeepNovo PointNovo Casanovo InstaNovo AdaNovo
Closest papers by meaning
- PepSpecBench: A Unified Evaluation Benchmark for Peptide Tandem Mass Spectrometry Prediction (NeurIPS 2026 · similarity 0.91)
- AdaNovo: Towards Robust \emph{De Novo} Peptide Sequencing in Proteomics against Data Biases (NeurIPS 2024 · similarity 0.91)
- Latent Imputation before Prediction: A New Computational Paradigm for De Novo Peptide Sequencing (ICML 2025 · similarity 0.90)
- ContraNovo: A Contrastive Learning Approach to Enhance De Novo Peptide Sequencing (AAAI 2024 · similarity 0.90)
- Universal Biological Sequence Reranking for Improved De Novo Peptide Sequencing (ICML 2025 · similarity 0.90)
- Curriculum Learning for Biological Sequence Prediction: The Case of De Novo Peptide Sequencing (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.