Deep Learning Enables Automatic Correction of Experimental HDX-MS Data with Applications in Protein Modeling

Area of Science:

• Biochemistry
• Analytical Chemistry
• Computational Biology

Background:

• Hydrogen-deuterium exchange mass spectrometry (HDX-MS) measures protein dynamics.
• Mass shifts in HDX-MS can deviate from expected values due to deuterium back and forward exchange.
• Accurate mass shift correction is crucial for advanced HDX-MS data processing and interpretation.

Purpose of the Study:

• To demonstrate the potential of deep neural networks for automatic correction of HDX-MS data.
• To evaluate the performance of an AI model in improving data fidelity across various analytical levels.
• To assess the utility of AI-corrected HDX-MS data for protein structure modeling.

Main Methods:

• Development and application of a multilayer perceptron (MLP) model.
• Training the MLP to learn the mapping between uncorrected and corrected HDX-MS mass shifts.
• Rigorous testing of the model at peptide, residue, and protein fold levels.

Main Results:

• The AI model successfully learned to correct mass shifts caused by deuterium exchange.
• Corrected data improved accuracy in calculating protection factors and identifying protein folds.
• The approach showed significant potential for enhancing HDX-MS data fidelity.

Conclusions:

• Deep neural networks offer a powerful tool for automated HDX-MS data correction.
• AI-driven correction enhances the reliability of HDX-MS for advanced applications, including protein modeling.
• Future online tools could predict corrected mass shifts, improving workflow efficiency and enabling retrospective data analysis.