Improving SWATH-MS analysis by deep-learning

Area of Science:

• Proteomics
• Analytical Chemistry
• Computational Biology

Background:

• Data-independent acquisition (DIA) mass spectrometry offers improved protein coverage and quantification.
• DIA success relies on high-quality spectral libraries, often generated via labor-intensive data-dependent acquisition (DDA).
• Theoretical spectral library generation algorithms exist but require optimization for specific DIA methods.

Purpose of the Study:

• To develop a deep-learning framework, dpSWATH, for generating accurate theoretical spectral libraries for SWATH-MS.
• To enhance the sensitivity and specificity of DIA data acquired on Q-TOF mass spectrometers.
• To improve protein identification rates in complex mixtures using SWATH-MS.

Main Methods:

• Development of a deep-learning model (dpSWATH) for predicting retention time and fragment ion intensity.
• Application of dpSWATH to generate theoretical spectral libraries for SWATH-MS experiments.
• Comparison of dpSWATH-generated libraries against traditional and library-free methods for protein identification.

Main Results:

• dpSWATH significantly increased the protein identification rate in SWATH-MS experiments.
• Theoretical libraries generated by dpSWATH demonstrated superior performance compared to existing algorithms.
• The method improved the quality of spectral data from Q-TOF mass spectrometers for SWATH analysis.

Conclusions:

• dpSWATH is a superior prediction framework for SWATH-MS, particularly for Q-TOF data.
• The deep-learning approach effectively addresses limitations in theoretical spectral library generation.
• This advancement facilitates more comprehensive and accurate proteomic analyses using DIA techniques.