Machine learning modeling of RNA structures: methods, challenges and future perspectives.

Area of Science:

• Molecular Biology
• Computational Biology

Background:

• RNA molecules are crucial for cellular functions, exhibiting dynamic three-dimensional structures that are essential for processes like riboswitching and epigenetic regulation.
• The dynamic and ensemble nature of RNA structures presents significant computational challenges for prediction, unlike advances seen in protein folding.

Approach:

• This review focuses on machine learning (ML) based methods for predicting RNA secondary and tertiary structures.
• It surveys common modeling strategies, including those incorporating thermodynamic principles.
• The review also discusses the limitations inherent in various design choices for these ML models.

Key Points:

• Machine learning offers promising approaches to tackle the complexity of RNA structure prediction.
• Understanding the dynamic ensemble of RNA structures is key to improving predictive accuracy.
• Thermodynamic principles can be integrated into computational models to enhance RNA structure prediction.

Conclusions:

• Current ML methods show potential but face challenges in accurately predicting the full spectrum of RNA structures.
• Future research should focus on developing more robust and accurate RNA structure prediction models by addressing existing limitations.
• Integrating diverse computational strategies and experimental data will be crucial for advancing the field.