De Novo Peptide Sequencing for Data-independent Acquisition (DIA) Using Deep Learning

thesis · 2025

thesis · 2025. Shiva Ebrahimi. In proteomics, de novo peptide sequencing from tandem mass spectrometry (MS/MS) data is a critical technology…
Date 2025-05-01
Type thesis
Publisher PhD thesis
Contribution algorithm
Supervisor Xuan Guo
DOI 10.12794/metadc2443151
Citations (OpenAlex) 0

Abstract

In proteomics, de novo peptide sequencing from tandem mass spectrometry (MS/MS) data is a critical technology for discovering novel peptide and protein sequences. Tandem mass spectrometry (MS/MS) stands as the predominant high-throughput technique for comprehensively analyzing protein content within biological samples. In recent years, substantial strides have been made in Data Independent Acquisition (DIA) strategies, facilitating impartial and non-targeted fragmentation of precursor ions. The DIA-generated MS/MS spectra present a formidable obstacle due to their inherent high multiplexing nature. This dissertation presents three deep-learning models designed to overcome the challenges of identifying novel peptide sequences directly from DIA spectra without relying on prior knowledge. First, we introduce ActiveNovo_DIA, an innovative framework that integrates the DeepNovo-DIA model with active learning algorithms. By employing various acquisition functions to selectively target the most informative spectra at each training iteration—rather than relying on random selection—this approach enhances the learning process and ultimately improves de novo sequencing performance. Next, we propose Transformer-DIA, a transformer-based de novo sequencing model specifically designed for DIA data. This model surpasses state-of-the-art (SOTA) methods by effectively addressing the inherent limitations of existing approaches. Its key contribution is expanding the spectrum encoder block, which enables the model to learn essential DIA spectral features more effectively. Finally, we extend the Transformer-DIA model by integrating a diffusion mechanism, leading to the development of DiffNovo. This enhancement not only improves model performance but also eliminates the need to compute high-dimensional features, addressing the limitations present in the original model. Our results demonstrate significant improvements over existing SOTA methods. Given the findings of two tools alongside three advanced models and their superior performance over current approaches, this dissertation contributes to advancing de novo peptide sequencing for DIA data, opening new avenues for proteomics research.

Authors

  1. Shiva Ebrahimi · University of North Texas

Methods and tools

  • AL-DeepNovo-DIA: 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.
  • DiaTrans: Transformer for DIA
  • DiffNovo-DIA: Transformer-diffusion model for DIA de novo peptide sequencing: DIA-side companion to DiffNovo. From Shiva Ebrahimi’s PhD thesis (UNT, 2025); a PyPI package exists (diffnovo-dia v0.1.2) but the GitHub repo is currently empty and no standalone paper has been published.

Cites (36)

Seen in the charts

Back to the full map

Back to top