AI Model Decodes Complex 'Knots' in RNA to Spot Disease Risks

While we often picture RNA as a simple linear messenger, it frequently folds into intricate three-dimensional structures. Among the most fascinating are RNA G-quadruplexes (rG4s)—dense, four-stranded knots that play a vital role in regulating gene expression. Until now, predicting how genetic mutations affect these structures has been a significant challenge.

A new tool called G4mer changes the landscape. This RNA language model utilises advanced computational modelling to predict rG4 formation and classify their subtypes with greater accuracy than previous methods. The study highlights that factors such as sequence length and the motifs flanking the structure are critical for accurate prediction.

Crucially, the team applied G4mer to the 5' untranslated regions—parts of the genome that do not code for protein directly—of genes associated with breast cancer. The model successfully identified specific genetic variants that disrupt rG4 formation, subsequently altering gene expression. This validates the importance of looking beyond standard coding regions when searching for disease drivers. By bridging computational predictions with experimental validation, G4mer offers a powerful new way to analyse the functional impact of non-coding variants.