BindRNAgen: Protein-binding RNA Sequence Generation Using Latent Diffusion Models.

Area of Science:

• Computational Biology
• RNA Biology
• Drug Discovery

Background:

• RNA-binding proteins (RBPs) regulate gene expression; their malfunction is linked to diseases.
• Developing synthetic RNA to modulate RBP activity is a potential therapeutic avenue.
• Current methods face challenges due to experimental screening inefficiency and limited computational model generalization.

Purpose of the Study:

• To introduce BindRNAgen, a novel hybrid computational framework for designing RNA molecules that bind specific RBPs.
• To enable the generation of functional RNA sequences using only the RBP sequence as input.
• To advance therapeutic strategies by improving the design of RNA-based modulators of RBP activity.

Main Methods:

• A hybrid framework combining a variational autoencoder (VAE) and a conditional latent diffusion model (LDM).
• Pretraining a VAE on RBP-binding RNA sequences from 168 eCLIP-seq datasets to learn sequence specificity.
• Utilizing an LDM conditioned on protein language model embeddings to generate novel RNA sequences.

Main Results:

• BindRNAgen generates RNA sequences with biophysical properties comparable to natural RBP-binding RNAs for known RBPs.
• The framework outperforms existing computational benchmarks in predicting RBP-binding RNA sequences.
• Generated RNA sequences exhibit high predicted binding scores, further validated by in silico docking and molecular dynamics simulations.

Conclusions:

• BindRNAgen offers a powerful computational approach for designing synthetic RNA molecules targeting specific RBPs.
• The framework shows promise for therapeutic applications by enabling efficient and accurate RNA design.
• While generalization to novel RBPs requires further investigation, the model demonstrates significant potential in RBP-RNA interaction prediction and design.