Illumina’s SpliceAI2 predicts RNA transcripts from DNA sequence
SpliceAI2 expands splice prediction to junctions and complete transcript isoforms. Illumina reports gains on research datasets, but the work is a non-peer-reviewed preprint and access runs through its BioInsight software.
2 min read
Illumina announced SpliceAI2 on October 8, 2026, a genomic model that predicts how DNA sequence affects RNA splicing. The update expands on the original SpliceAI, which focused on whether a splice site appears at a given position: SpliceAI2 also predicts which splice sites connect and which full-length transcript isoforms may result.
Illumina says the model was trained on RNA-sequencing data from 314,745 samples across ten species, covering more than 46 million splice junctions, plus 330 long-read samples from ENCODE. In the company’s evaluation, it reconstructed the most common transcript for previously unseen genes 82% of the time, compared with 78% for a short-read-only training setup.
On a Genomics England cohort of 7,504 people, Illumina reports that SpliceAI2 identified 17% more disease-associated variants than other tested splicing models at matched confidence thresholds. The company also reports a 34% improvement in splice-site-usage quantification versus the next-best model in GTEx analyses. An independent Rare Daily report confirms the announced figures while correctly attributing them to Illumina and its preprint.
These are promising research results, not proof of clinical benefit. The technical overview and performance claims come from Illumina’s BioInsight AI Lab, and the preprint is not peer reviewed. Illumina says researchers can access the model through BioInsight applications, including DRAGEN Annotation, Emedgene and Illumina Connected Insights. It has not announced downloadable weights or a permissive model license.
The company labels the material “For Research Use Only” and says it is not for diagnostic procedures. Researchers will need independent validation—especially across populations and clinical settings—before treating the reported benchmark gains as evidence of real-world diagnostic value.
Sources: Illumina’s release · Technical overview and evaluation details · Rare Daily coverage
Conversation
0 approved commentsNo comments yet. Start a thoughtful conversation.
Join the conversation.
Sign in with ChatGPT to leave a comment or rate this article.