Struct2SeQ: RNA inverse folding with Deep Q-Learning.

Area of Science:

• Biochemistry
• Computational Biology
• Synthetic Biology

Background:

• RNA secondary structure dictates function in biological and therapeutic applications.
• Designing RNA sequences for specific structures, particularly complex pseudoknots, is a significant challenge.

Purpose of the Study:

• To develop an AI framework for generating RNA sequences that fold into desired secondary structures.
• To incorporate SHAPE reactivity constraints for enhanced chemical validity.

Main Methods:

• Utilized a reinforcement learning framework, specifically deep Q-learning, for RNA sequence generation.
• Formulated RNA design as a sequential decision-making process.
• Incorporated SHAPE-informed rewards to guide sequence generation.

Main Results:

• Struct2SeQ successfully generated RNA sequences matching target secondary structures and SHAPE profiles.
• The framework significantly outperformed human designs and other automated methods in pseudoknot design challenges.
• Generated sequences demonstrated improved chemical validity and explored a broader sequence space.

Conclusions:

• Reinforcement learning offers a powerful approach for complex RNA design.
• Struct2SeQ advances the engineering of RNA molecules with precise structures and functions.
• This method holds promise for future RNA-based therapeutics and biotechnology.