San Diego — October 8, 2026
What happened: Illumina introduced SpliceAI2, a genomic AI model from its BioInsight AI Lab that predicts how genetic variants alter RNA splicing — identifying 17% more disease-relevant variants than competing models in rare-disease research data. Why it matters: The first SpliceAI became a workhorse of clinical genomics; its successor is aimed at the variants that today’s sequencing keeps missing.
What SpliceAI2 does
SpliceAI2 takes a DNA sequence as input and predicts which parts of a gene get used as splice sites, which sites connect through splice junctions, and which full-length RNA transcript isoforms a cell will produce. That matters because splice-altering variants — changes that disrupt how RNA is assembled — account for a meaningful share of genetic disease, and many are hard to detect with standard exome sequencing, which typically misses variants more than 50 base pairs deep into intronic regions.
The model was trained end-to-end on 314,745 RNA sequencing samples spanning humans and nine other mammalian species, covering more than 46 million observed splice junctions — a dataset Illumina says is more than 100 times larger than its predecessor’s.
The details
In an analysis of 7,504 participants from the Genomics England 100,000 Genomes Project, variants prioritized by SpliceAI2 were significantly enriched in phenotype-matched disease genes, identifying 17% more disease-associated variants than any other tested splicing model at matched confidence thresholds. Illumina says the model found 33% more disease-relevant splice variants than the original SpliceAI, and 66% more at high confidence. Across three independent benchmarks, the company reports SpliceAI2 outperformed SpliceAI, Pangolin, and Google DeepMind’s AlphaGenome — with the AlphaGenome comparisons run independently by collaborators at the University of Oxford.
SpliceAI2 joins PromoterAI and PrimateAI-3D in Illumina’s suite of genomic AI models covering splice, promoter, and missense variant effects; the company says the three together now let researchers identify up to twice as many variants with predicted biological impact.
The model is accessible through Illumina’s BioInsight Platform applications, including DRAGEN Annotation, Emedgene, and Illumina Connected Insights. Source code, trained weights, and precomputed predictions for 4 billion single-nucleotide variants are available on GitHub and Hugging Face for academic and non-commercial research use, with commercial licensing handled separately.
Why it matters
The original SpliceAI, released in 2019, has been cited in more than 3,400 publications and is baked into ClinGen’s splice variant interpretation standards — so a genuinely better successor lands directly in real diagnostic workflows. The practical effect is fewer “variants of uncertain significance” in rare-disease cases and hereditary cancer testing, where families can wait years for an answer. The open release of code and weights also keeps the genomics community honest: anyone can run the benchmarks themselves. We have not independently verified the benchmark claims, which so far come from the company’s own manuscript, but the Oxford-run AlphaGenome comparison is a credibility signal worth noting.
Frequently asked questions
What does SpliceAI2 predict? How DNA sequence variants affect RNA splicing — which splice sites are used, how they connect, and which RNA transcripts result.
How is it different from the original SpliceAI? A training set more than 100 times larger, end-to-end transcript-level prediction, and roughly half again as many disease-relevant variants found at matched confidence.
Is it available to use? Yes — through Illumina’s BioInsight Platform tools, and as open code and weights on GitHub and Hugging Face for academic and non-commercial research. It needs a CUDA-capable GPU.
Does it diagnose disease? No — it prioritizes variants for researchers and clinicians interpreting sequencing data; it does not itself make clinical diagnoses.
Sources: Unite.AI; Illumina announcement