Complementing machine learning-based structure predictions with native mass spectrometry.

Area of Science:

• Structural Biology
• Computational Biology
• Biochemistry

Background:

• Machine learning (ML) algorithms, including AlphaFold2 (AF2) and RoseTTa Fold, enable accurate prediction of protein structures.
• ML-based structure predictions for protein complexes often require experimental validation due to reliance on user input.
• Mass spectrometry (MS) is a powerful technique for analyzing protein modifications, interactions, and structural dynamics.

Purpose of the Study:

• To investigate the complementary capabilities of native MS and ML for protein structure prediction.
• To demonstrate how MS can reveal structural features not detectable by ML algorithms alone.

Main Methods:

• Utilized three distinct protein systems for experimental analysis.
• Employed native mass spectrometry (MS) to probe protein complex structures and interactions.
• Compared MS findings with predictions generated by ML-based structure prediction tools like AF2.

Main Results:

• Native MS experiments successfully identified structural features related to ligand interactions, homology models, and point mutations.
• These specific structural details were found to be undetectable by AF2 predictions alone.
• Demonstrated the ability of MS to uncover critical information beyond the scope of current ML structure prediction.

Conclusions:

• Native MS is a valuable experimental technique for validating and refining ML-based protein structure predictions.
• The integration of ML and MS provides a more comprehensive approach to structural modeling.
• Combining computational predictions with experimental MS data yields more accurate structural models at various scales.