Extracting Residue Solvent Exposure from Covalent Labeling Data with Machine Learning: A Hybrid Approach for Protein...

Area of Science:

• Structural biology
• Computational biology
• Biochemistry

Background:

• Hydroxyl radical protein footprinting (HRPF) combined with mass spectrometry provides data on residue solvent exposure and protein topology.
• Experimental HRPF data is sparse and requires computational interpretation for structural insights.
• Existing protein structure prediction methods like AlphaFold2 (AF2) utilize machine learning (ML) but integrating HRPF data is challenging.

Purpose of the Study:

• To develop a hybrid computational method integrating experimental HRPF data with ML for enhanced protein structure prediction.
• To overcome the limitations of sparse HRPF data and the abstract nature of ML networks in structure prediction.

Main Methods:

• A hybrid approach was developed using a light gradient boosting machine to predict residue solvent accessibility from HRPF data.
• These solvent accessibility predictions were then used to refine protein structure predictions with the Rosetta algorithm.

Main Results:

• The hybrid method successfully predicted atomic-level detail for all four proteins in the benchmark set.
• The approach demonstrates the feasibility of using ML with HRPF data for accurate protein structure prediction.

Conclusions:

• Combining experimental HRPF data with ML offers a powerful strategy for improving protein structure prediction.
• This hybrid method enhances the accuracy and detail of computational protein modeling.