Prosit-TMT: Deep Learning Boosts Identification of TMT-Labeled Peptides

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Accurate prediction of peptide properties is crucial for mass spectrometry-based proteomics.
• Existing prediction models primarily focus on unlabeled peptides, limiting their application in multiplexed analyses.
• Tandem Mass Tags (TMT) enable multiplexed quantification but require specialized prediction tools.

Purpose of the Study:

• To develop an advanced deep learning model for predicting retention time and fragment ion intensities of TMT-labeled peptides.
• To create an extensive resource of synthetic TMT-labeled peptides for model training and validation.
• To improve the identification and quantification of peptides in complex proteomic datasets.

Main Methods:

• Generation of a comprehensive dataset of synthetic TMT-labeled peptides.
• Extension of the deep learning model Prosit to incorporate TMT labeling.
• Training and validation of the Prosit-TMT model using the synthetic peptide resource.
• Application of the Prosit-TMT model to reanalyze existing TMT-labeled proteomic datasets.

Main Results:

• The Prosit-TMT model demonstrates high accuracy in predicting retention time and fragment ion intensities for TMT-labeled peptides.
• The model supports both Collision-Induced Dissociation (CID) and Higher-energy Collisional Dissociation (HCD) fragmentation.
• The model integrates predictions for ion trap and Orbitrap mass analyzers within a single framework.
• Reanalysis of published datasets using Prosit-TMT yielded substantial additional proteomic information.

Conclusions:

• Prosit-TMT significantly enhances the analysis of TMT-labeled proteomic data.
• The model facilitates more confident peptide identification and quantification in multiplexed experiments.
• Application of Prosit-TMT revealed daily cyclical protein expression patterns in human breast milk, potentially impacting infant development.