Learning the Fastest RNA Folding Path Based on Reinforcement Learning and Monte Carlo Tree Search

Area of Science:

• Computational Biology
• Molecular Biology
• Artificial Intelligence

Background:

• RNA molecules require specific structures for biological functions.
• Understanding RNA folding mechanisms is crucial but challenging.
• Current methods primarily focus on structure prediction, not folding dynamics.

Purpose of the Study:

• To investigate the fastest folding paths of RNA secondary structures.
• To develop a novel computational approach for studying RNA folding mechanisms.
• To apply deep reinforcement learning to RNA folding prediction.

Main Methods:

• Developed 2dRNA-Fold, a deep reinforcement learning algorithm.
• Utilized a neural network combined with Monte Carlo tree search.
• Applied the algorithm to predict folding paths for short and long RNA molecules.

Main Results:

• Identified interesting features in the fastest folding paths of tested RNA molecules.
• Successfully trained 2dRNA-Fold on the bpRNA dataset for RNA secondary structure prediction.
• Demonstrated the algorithm's capability to predict folding pathways.

Conclusions:

• 2dRNA-Fold offers a new perspective on RNA folding dynamics.
• The predicted fastest folding paths may differ from free-energy-based predictions.
• This approach advances the study of RNA folding mechanisms.