Predicting the Flexibility Profile of Ribosomal RNAs

Area of Science:

• Biomolecular structure and dynamics
• Computational biology
• Nucleic acid research

Background:

• Biomolecular flexibility, quantified by the B-factor, is vital for function.
• Protein B-factor prediction is established, but methods for ribosomal RNAs (rRNAs) are underdeveloped.
• rRNAs are functionally analogous to proteins, serving as structural scaffolds and catalysts.

Purpose of the Study:

• To develop quantitative structure-flexibility relationship (QSFR) models for predicting rRNA B-factors.
• To compare sequence-based and structure-based prediction approaches.
• To evaluate machine learning methods for rRNA flexibility prediction.

Main Methods:

• Applied linear and nonlinear machine learning algorithms: Partial Least Squares Regression (PLS), Least Squares Support Vector Machine (LSSVM), and Gaussian Process (GP).
• Utilized both primary sequence and advanced structural data for model development.
• Rigorously examined model performance and reliability against established protein B-factor models.

Main Results:

• rRNA B-factor prediction accuracy is comparable to that achieved for proteins.
• Structure-based approaches significantly outperform sequence-based methods for rRNA B-factor modeling.
• rRNA flexibility is predominantly influenced by local nonbonding potential landscapes (electrostatic and van der Waals forces).

Conclusions:

• Machine learning effectively predicts rRNA B-factors, achieving protein-level accuracy.
• Structural information is superior to sequence information for modeling rRNA flexibility.
• Local nonbonding interactions are key determinants of rRNA flexibility.