Modern genome and exome sequencing can identify thousands of genetic variants, including many located within introns, the stretches of DNA that do not directly code for proteins. While these regions were once thought to be unimportant, researchers now know that changes within introns can disrupt RNA splicing, the process that removes introns and joins coding regions together to produce functional messenger RNA. Determining which intronic variants are truly harmful, however, remains a major challenge.

Researchers at the University Hospital of Montpellier in France investigated whether combining multiple computational prediction tools with functional RNA analysis could improve the interpretation of intronic variants found in patients with neurodegenerative diseases.

The team examined nine genetic variants whose effects on RNA splicing were uncertain. These included deep intronic variants located far from normal splice sites, non-canonical splice-site variants, and one canonical splice-site variant associated with an unusual clinical presentation. Several state-of-the-art deep learning and motif-based prediction tools were used to estimate how each variant might alter RNA splicing.

The computational predictions suggested a variety of possible mechanisms. Some variants were predicted to disrupt existing donor splice sites, while others appeared to create new cryptic splice sites that could cause abnormal RNA processing. One variant was predicted to activate a deep intronic splicing enhancer, a regulatory sequence that encourages inappropriate inclusion of intronic regions within the mature RNA transcript.

To determine which predictions were correct, the researchers performed transcript analysis using RNA isolated from patient blood samples. Functional RNA analysis confirmed abnormal splicing in eight of the nine cases. Importantly, every computational method failed to correctly predict at least one variant. Deep learning models occasionally missed donor splice-site loss or cryptic splice-site activation, while motif-based approaches sometimes produced false positive predictions or overlooked functional splice sites.

The investigators also used minigene experiments to further examine one unusual variant. These experiments supported a role for the splicing factor SRSF2 in promoting inclusion of a pseudoexon, demonstrating how experimental validation can uncover regulatory mechanisms that prediction algorithms alone may miss.

The findings highlight that no single computational tool is sufficient for interpreting all splicing variants. Instead, combining multiple prediction methods with functional RNA analysis provides a more reliable strategy for determining whether a genetic variant is disease causing. As RNA sequencing and transcript-level analyses become increasingly integrated into clinical genetics, this combined approach may improve diagnostic accuracy for patients with inherited neurological disorders and other genetic diseases.