ResNeXt-Based Rescoring Model for Proteoform Characterization in Top-Down Mass Spectra

Area of Science:

• Proteomics
• Computational Biology
• Mass Spectrometry

Background:

• Accurate proteoform identification is crucial in top-down proteomics.
• Protein primary structure variations create diverse proteoforms, complicating analysis.
• Lack of a reference set hinders standardization and accuracy.

Purpose of the Study:

• To develop and validate PrSMBooster, a deep learning model for improving proteoform characterization.
• To enhance the accuracy of proteoform spectrum match (PrSM) scoring.

Main Methods:

• Introduced PrSMBooster, a ResNeXt-based deep learning model.
• Utilized an ensemble approach integrating logistic regression, XGBoost, decision tree, and support vector machine.
• Input basic and latent PrSM features into the ResNeXt model for rescoring.

Main Results:

• PrSMBooster increased the number of identified proteoform spectrum matches (PrSMs) at a 1% false discovery rate across 47 datasets.
• Demonstrated improved accuracy in PrSM scoring compared to the TopPIC algorithm.
• Showcased strong generalization ability across diverse mass spectrometry datasets.

Conclusions:

• PrSMBooster significantly enhances the accuracy and scope of proteoform characterization in top-down proteomics.
• The model offers a robust solution for improving mass spectrometry data analysis.
• PrSMBooster's generalization ability makes it valuable for various proteomic studies.