Machine learning deciphers structural features of RNA duplexes measured with solution X-ray scattering.

Area of Science:

• Biophysics
• Structural Biology
• Computational Biology

Background:

• Solution X-ray scattering, including small-angle (SAXS) and wide-angle (WAXS) techniques, provides insights into macromolecular structures.
• WAXS extends resolution below 10 Å, offering critical details for biological function but presents interpretation challenges due to solvent contributions.
• Double-stranded RNA (dsRNA) duplex conformations are sensitive to salt and sequence, influencing RNA foldability and function.

Purpose of the Study:

• To develop and apply machine learning (ML) methods for interpreting WAXS data of dsRNA.
• To predict key structural parameters of dsRNA duplexes using WAXS profiles.
• To establish a framework integrating theoretical models with experimental scattering data for detailed macromolecular structure analysis.

Main Methods:

• Utilized supervised machine learning, specifically extreme gradient boosting (XGBoost), trained on molecular dynamics-generated models.
• Treated WAXS profiles as 1D image features for ML model input.
• Applied ML models to predict structural parameters of dsRNA duplexes based on their scattering profiles.

Main Results:

• The ML models successfully identified specific scattering angles and regions correlating with distinct structural parameters.
• XGBoost effectively predicted key structural parameters of dsRNA duplexes from WAXS data.
• Demonstrated the capability of ML to extract high-resolution structural information from solution scattering experiments.

Conclusions:

• Machine learning strategies provide a powerful new framework for analyzing solution scattering data.
• This approach enhances the extraction of highly relevant structural information from WAXS experiments on biological macromolecules.
• The integration of ML with experimental data and theoretical models advances macromolecular structure determination.