Symposium Sessions
Day 1, June 10（Wed.） 13:50-14:08 Room C (4F 413)
- 1C-S1-1350
Deep learning in proteomics mass spectrometry data analysis
(Fudan Univ)
oLiang Qiao
Deep learning has driven significant advances in proteomics, enabling applications such as MS/MS spectrum prediction, de novo peptide sequencing, and retention time prediction. In 2020, we introduced DeepDIA, a deep learning tool that generates in silico spectral libraries for data-independent acquisition (DIA) analysis. It produces instrument-specific libraries comparable to experimental ones and supports library generation directly from protein sequence databases using peptide detectability prediction. In 2023, we developed DeepFLR, a framework for controlling false localization rates (FLR) in phosphoproteomics. It includes a deep learning-based phosphopeptide MS/MS prediction module and a target-decoy FLR assessment module, achieving higher prediction accuracy and improved phosphosite localization across diverse datasets. Most recently, we presented DeepGP, a hybrid Transformer and graph neural network framework for predicting glycopeptide MS/MS spectra and retention time. DeepGP accurately models experimental data, distinguishes similar and isomeric glycopeptides, and when combined with database searching, significantly enhances glycopeptide detection sensitivity.