Related Experiment Video
Updated: May 13, 2025
09:47
Author Spotlight: Advancing Alzheimer's Research – Exploring Early Detection and Multi-Omics Approaches
Published on: December 15, 2023
910
DIA-BERT: pre-trained end-to-end transformer models for enhanced DIA proteomics data analysis
Zhiwei Liu1,2,3, Pu Liu4, Yingying Sun1,2,3
1Affiliated Hangzhou First People's Hospital, State Key Laboratory of Medical Proteomics, School of Medicine, Westlake University, Hangzhou, Zhejiang Province, China.
Nature Communications
|April 14, 2025
Summary
DIA-BERT, a new AI tool, significantly improves protein and peptide precursor identification in data-independent acquisition mass spectrometry (DIA-MS) proteomics analysis. It enhances quantitative accuracy in cancer sample datasets.
Area of Science:
- Proteomics
- Mass Spectrometry
- Bioinformatics
Background:
- Data-independent acquisition mass spectrometry (DIA-MS) is crucial for quantitative proteomics.
- Existing DIA-MS analysis tools have limitations in identifying and quantifying proteins.
Purpose of the Study:
- To introduce DIA-BERT, an AI-powered software for analyzing DIA-MS data.
- To evaluate DIA-BERT's performance against existing tools like DIA-NN.
Main Methods:
- Developed DIA-BERT using a transformer-based AI model.
- Trained the identification model on over 276 million peptide precursors from DIA-MS files.
- Trained the quantification model on 34 million peptide precursors from synthetic DIA-MS files.
Main Results:
- DIA-BERT achieved a 51% increase in protein identifications compared to DIA-NN.
- DIA-BERT identified 22% more peptide precursors on average across five human cancer datasets.
- Demonstrated high quantitative accuracy in complex proteomic samples.
Conclusions:
- Leveraging pre-trained AI models and synthetic datasets significantly enhances DIA-MS data analysis.
- DIA-BERT represents a substantial advancement in DIA proteomics quantification and identification.
- The findings highlight the potential of AI in accelerating proteomic discoveries.
