# Modeling RNA duplex dynamics with Gibbs sampling enhances base-pair prediction accuracy and reveals structural activity profiles

## TL;DR

This paper introduces a new method for predicting RNA structures using Gibbs sampling, improving accuracy and providing insights into RNA interactions.

## Contribution

The novel contribution is the application of Gibbs sampling to model RNA duplex dynamics, enhancing prediction accuracy and providing structural activity profiles.

## Key findings

- MC-DuplexFold improves base-pair prediction accuracy when combined with other RNA structure prediction algorithms.
- Approximate methods like RIsearch and Sfold outperform exact methods in structural prediction accuracy.
- MC-DuplexFold provides structural activity statistics useful for modeling miRNA interactions.

## Abstract

The RNA secondary (2D) structure prediction problem consists in determining the set of base pairs that form within an RNA molecule from its sequence. A related task is the RNA hybridization problem, where two RNA strands interact to form a duplex. Thermodynamics-based methods typically rely on experimentally determined energy parameters to compute minimum free energy structures for both single-stranded RNAs and duplexes. Through the Boltzmann distribution, these parameters can be used to estimate base-pairing probabilities. Here, we leverage these probabilities to simulate RNA:RNA interaction dynamics. Inspired by the Ising model, we apply Gibbs sampling to model the stochastic formation and disruption of base pairs over time in RNA duplexes, ultimately deriving a consensus structure. The resulting method, MC-DuplexFold (mcdf), enhances base-pair prediction accuracy when integrated with other RNA 2D structure prediction algorithms. Through benchmarking, we reaffirm the previously observed trend that approximate or heuristic methods, such as RIsearch and Sfold, outperform exact methods like RNAcofold and DuplexFold in structural prediction accuracy. Additionally, mcdf provides structural activity statistics that can be incorporated into the modeling of miRNA primary transcripts, precursors, and target interactions, thereby refining predictions of miRNA:mRNA duplex dynamics.