Illumina Launches AI Tool to Help Researchers Find Overlooked Rare Disease Variants

Rare Daily Staff

Illumina has introduced a new artificial intelligence tool designed to help researchers identify genetic variants that may contribute to rare diseases but escape detection by existing methods.

The company said the new SpliceAI2 model identified 17 percent more disease-relevant splicing variants than competing prediction tools in an analysis of rare disease data. The findings are described in a newly released preprint.

SpliceAI2 addresses a persistent challenge in rare disease research: determining which genetic changes matter. Sequencing can identify variants in a person’s DNA, but determining whether those variants disrupt a gene’s function and could explain a disease is a separate task. Some are classified as variants of uncertain significance because their effects remain unclear.

SpliceAI2 focuses on variants that affect RNA splicing, a step in the process cells use to make proteins. During splicing, cells remove certain sections of an RNA transcript and join the remaining sections. Genetic variants that interfere with this process can alter the resulting transcript and disrupt gene or protein function.

Some variants activate “cryptic” splice sites—normally unused locations where RNA is cut and joined that can be difficult to identify using traditional approaches. Illumina said SpliceAI2 is designed to help researchers detect and interpret these variants, including by predicting how splicing differs among cell types.

“Today, we are equipping researchers with the next generation of technology to help understand the underlying causes of disease,” said Kyle Farh, vice president of Illumina’s BioInsight AI Lab.

In an analysis of whole-genome and phenotype data from 7,504 participants in Genomics England’s 100,000 Genomes Project, the model identified 17 percent more disease-relevant splice variants than the next-best model at a matched confidence threshold, according to the preprint.

In a separate analysis of matched whole-genome and RNA sequencing data from the National Institutes of Health’s Genotype-Tissue Expression resource, SpliceAI2 improved the quantification of splice-site usage by 34 percent compared with the next-best model, Illumina said.

Illumina also reported that the model’s predictions aligned more closely with changes in protein abundance observed in UK Biobank data and better captured the effects of splice variants across several population datasets.

SpliceAI2 was trained on a dataset more than 100 times larger than the one used for the original SpliceAI. Illumina reported that the new model detected more disease-relevant variants while reducing the number requiring review—a potential benefit for researchers evaluating large numbers of genetic findings.

The new model joins two other Illumina tools, PromoterAI and PrimateAI-3D, that assess different types of genetic variation. Illumina said the three models together allow researchers to identify up to twice as many variants with predicted biological effects.