| 引用本文: | 邱宁宁,张航,柳楠.基于Transformer的多种蛋白质从头肽测序[]技术在基准数据集上的评测[J].生物信息学,2026,24(2):101-113. |
| QIU Ningning,ZHANG Hang,LIU Nan.Evaluation of multiple protein de novo peptide sequencing technologiesbased on transformer on benchmark datasets[J].Chinese Journal of Bioinformatics,2026,24(2):101-113. |
|
| 摘要: |
| 从头肽测序作为质谱分析中解析蛋白质结构与功能的关键技术,近年来深度学习模型尤其是 Transformer 架构的融入,展现出巨大发展潜力。本文旨在聚焦基于 Transformer 的多种蛋白质从头肽测序技术,通过在基准数据集上的评测,为该领域的研究提供全面且深入的参考。从众多基于Transformer架构的模型中,选取Casanovo、CasanovoV2、Transformer-DIA、π-HelixNovo、PowerNovo 以及π-PrimeNovo,在基准数据集上开展对比实验。从肽测序准确率、运行效率等多个维度,对各模型的性能表现差异展开深入剖析,详细分析各模型在处理质谱数据、预测肽序列过程中所采用的不同策略。与传统的肽测序方法相比,Transformer模型的并行处理和全局依赖建模能力为肽测序带来了革新,通过降低复杂度、提高精确度和效率,解决了传统方法中的错误级联问题,为蛋白质组学研究提供了不可或缺的价值。通过对多种基于 Transformer 的蛋白质从头肽测序技术在基准数据集上的评测与分析,为读者提供一个全面的视角,明确了这些模型在推动蛋白质组学研究中的重要作用,同时也为后续研究指明了方向。 |
| 关键词: 质谱分析 从头肽测序 Transformer 蛋白质组学 |
| DOI：10.12113/202411012 |
| 分类号:Q51 |
| 文献标识码:A |
| 基金项目:国家自然基金青年项目(No.61902221);2021年山东省研究生教育教学改革研究项目(No.SDYJG21173). |
|
| Evaluation of multiple protein de novo peptide sequencing technologiesbased on transformer on benchmark datasets |
|
QIU Ningning,ZHANG Hang,LIU Nan
|
|
(School of Computer Science and Technology, Shandong Jianzhu University,Jinan 250101,China)
|
| Abstract: |
| De novo peptide sequencing, as a key technology for analyzing the structure and function of proteins in mass spectrometry analysis, has shown great development potential in recent years with the integration of deep learning models, especially the Transformer architecture. This study aims to focus on various de novo peptide sequencing technologies for proteins based on the Transformer, and through evaluations on benchmark datasets, provide a comprehensive and in-depth reference for research in this field. In this study, from among numerous models based on the Transformer architecture, Casanovo, CasanovoV2, Transformer-DIA, π-HelixNovo, PowerNovo, and π-PrimeNovo were selected to conduct comparative experiments on benchmark datasets. The differences in the performance of each model were deeply analyzed from multiple dimensions such as peptide sequencing accuracy and operation efficiency. The different strategies adopted by each model in processing mass spectrometry data and predicting peptide sequences were analyzed in detail. Compared with traditional peptide sequencing methods, the parallel processing and global dependency modeling capabilities of Transformer models have brought innovation to peptide sequencing. By reducing complexity, improving accuracy and efficiency, it has solved the error cascade problem in traditional methods, providing indispensable value for proteomics research. Through the evaluation and analysis of various Transformer-based de novo peptide sequencing techniques for proteins on benchmark datasets, this paper provides readers with a comprehensive perspective, clarifies the important role of these models in promoting proteomics research, and also points out the direction for subsequent research. |
| Key words: Mass spectrometry De novo peptide sequencing Transformer Proteomics |
