Photo credit: phys.org

New Method Enhances Understanding of RNA Structures

Researchers at the National Cancer Institute have unveiled a revolutionary method known as HORNET, designed to elucidate the three-dimensional topological arrangements of large and flexible RNA molecules. This innovative approach combines atomic force microscopy (AFM) with deep neural networks and unsupervised machine learning, allowing scientists to analyze individual RNA conformers in physiological conditions.

Human RNA molecules encompass various structural elements that play vital roles in biological processes. Traditional techniques, such as cryo-electron microscopy, typically necessitate highly uniform samples and extensive signal averaging. As a result, the analysis of large, flexible RNAs has long posed a significant challenge, as these molecules often adopt multiple conformations when dissolved in solution.

Currently, there is no comprehensive database that links RNA sequences with their corresponding three-dimensional structures. The successful methodologies that have been developed for proteins, like AlphaFold, do not extend to RNA, highlighting a critical void in the field of structural biology. The scarcity of RNA-focused deep-learning strategies underscores the complexities involved in generating reliable structural models.

The study titled “Determining structures of RNA conformers using AFM and deep neural networks,” which appears in Nature, introduces HORNET and outlines its significant potential for uncovering previously obscured structural features of large and flexible RNA molecules.

To validate HORNET, researchers gathered single-molecule AFM images of benchmark RNAs displayed in various conformations. They employed unsupervised machine learning and deep neural networks to establish connections between molecular topographies and their energy distributions.

Training and Testing of HORNET

The system was trained on a pseudo-structure database that encompassed a wide array of RNA folds and subsequently tested on several RNA samples exceeding 200 nucleotides in length, including RNase P RNA, a cobalamin riboswitch, a group II intron, and the HIV-1 Rev response element RNA. Various initial models were utilized, incorporating predicted structures and conformers derived from small-angle X-ray scattering data.

Results from test cases indicated that HORNET effectively reconstructed individual RNA conformations, with root-mean-square deviations—an indicator of how closely a calculated structure aligns with an established reference—often falling beneath the critical 7 Å threshold commonly used to validate significant structural features in large RNA molecules.

Further benchmark experiments, using both simulated and experimental AFM images, confirmed the robustness of integrating established constraints alongside AFM pseudo-potentials. Validation procedures illustrated that the diverse conformations of both RNase P RNA and HIV-1 Rev response element RNA could be observed at the single-molecule level, with accuracy estimations from the deep neural networks aligning closely with actual measurements from known structures.

Implications of HORNET in RNA Research

HORNET provides a solution to one of the major obstacles in RNA structural biology by offering a comprehensive, direct method to examine previously hard-to-detect RNA structures. This advancement holds significant promise for future research in various domains, including clinical studies, pharmaceuticals, and biotechnological applications.

More information: Maximilia F. S. Degenhardt et al, “Determining structures of RNA conformers using AFM and deep neural networks,” Nature (2024). DOI: 10.1038/s41586-024-07559-x

Source
phys.org