RNAGenScape: Property-Guided, Optimized Generation of mRNA Sequences with Manifold Langevin Dynamics

Area of Science:

• Computational Biology
• Bioinformatics
• Synthetic Biology

Background:

• Optimizing messenger RNA (mRNA) sequences is crucial for advanced therapeutics like vaccines and protein replacement.
• Current generative methods struggle with limited data and complex sequence-function relationships, often producing non-functional sequences.
• Ensuring biological viability is a key challenge in mRNA sequence design.

Purpose of the Study:

• To develop a novel framework, RNAGenScape, for generating property-optimized mRNA sequences.
• To address the limitations of existing generative methods by ensuring biological viability.
• To improve the efficiency and success rate of mRNA sequence generation.

Main Methods:

• RNAGenScape utilizes a property-guided manifold Langevin dynamics framework operating on a learned data manifold.
• It integrates an autoencoder with a property predictor to create a property-organized latent manifold.
• A denoising autoencoder and property-guided Langevin dynamics ensure optimization stays within biologically viable regions.

Main Results:

• RNAGenScape demonstrated significant improvements across three real-world mRNA datasets.
• The framework increased median property gain by up to 148% and success rate by up to 30%.
• Generated sequences maintained biological viability, outperforming existing generative approaches in efficiency.

Conclusions:

• RNAGenScape offers a robust solution for generating high-quality, biologically viable mRNA sequences.
• The property-guided manifold approach effectively navigates the complex sequence space.
• This framework advances mRNA sequence optimization for therapeutic applications.