VIRSE: a variational Bayesian framework for RNA structural ensemble inference

Area of Science:

• Computational Biology
• Structural Biology
• Biophysics

Background:

• RNA molecules exhibit dynamic structural heterogeneity, forming ensembles not captured by single-structure predictions.
• Chemical probing methods (DMS-MaPseq, SHAPE-MaP) provide single-molecule signals reflecting RNA structural diversity.
• Existing ensemble-inference methods like expectation-maximization (EM) face challenges with stability, scalability, and local optima for complex RNA data.

Purpose of the Study:

• To develop a robust and scalable computational framework for reconstructing RNA conformational ensembles from chemical probing data.
• To overcome the limitations of existing ensemble-inference approaches, particularly for high-dimensional and sparse mutation matrices.
• To enable accurate and efficient analysis of RNA structural heterogeneity and dynamics.

Main Methods:

• Developed VIRSE, a variational Bayesian framework utilizing coordinate ascent variational inference.
• Evaluated VIRSE using extensive simulations, including mechanism-informed mutation simulations mimicking DMS-MaP-seq.
• Tested VIRSE on idealized Bernoulli-mixture datasets and real-world experimental datasets.

Main Results:

• VIRSE demonstrated superior ensemble separation and cluster identifiability compared to EM across various simulation conditions.
• The framework showed improved stability, resolved low-abundance states, and scaled effectively to thousands of nucleotide positions.
• VIRSE successfully reconstructed biologically meaningful RNA conformational ensembles from experimental data.

Conclusions:

• VIRSE provides an efficient, scalable, and noise-robust solution for reconstructing RNA conformational mixtures from chemical probing data.
• The variational Bayesian approach overcomes limitations of traditional EM methods for complex RNA structural analysis.
• VIRSE accurately captures RNA structural heterogeneity, offering valuable insights into RNA function and dynamics.