Skip to content

SpliceAI

Summary

SpliceAI is a deep learning-based tool developed by Illumina to predict splice sites and the functional impact of genomic variants on pre-mRNA splicing. By evaluating flanking nucleotide contexts, the tool calculates a delta score to predict donor and acceptor gains or losses. This enables the computational identification of putatively pathogenic cryptic splice variants that reside outside canonical splice sites.

Overview

SpliceAI uses a deep residual network to evaluate pre-mRNA sequences and predict splicing motifs (splice donor and splice acceptor sites). It handles variant annotation by predicting how single nucleotide variants (SNVs) or indels disrupt splicing.

In the large-scale blood lipids ExWAS study by Koyama et al. (2026), SpliceAI was applied to evaluate the splicing impact of all identified variants in a cohort of 1,158,017 individuals. Variants with a Delta Score (DS) > 0.8 were classified as putatively cryptic splice variants and reclassified as predicted loss of function (pLoF) alleles, demonstrating equivalent phenotypic effect sizes to canonical pLoF variants.

Usage in Lipid ExWAS

  • Threshold: A delta score (DS) > 0.8 was used to define cryptic splice variants.
  • Variant Classification: Non-pLoF variants exceeding this threshold were reclassified as pLoF.
  • Findings: The study identified 15 exome-wide significant (EWS) cryptic splice variants, including a highly significant splice donor 5th base variant in APOA5 (rs200831171) associated with increased triglycerides.
  • Cryptic Splice Variants — the non-canonical splice effects identified using SpliceAI delta scores.
  • Missense Score — an ensembled deleteriousness metric used alongside splice-impact annotation in the lipid ExWAS.

Citations

  • [1] Jaganathan, K. et al. (2019). Predicting splicing from primary sequence with deep learning. Cell, 176(3), 535–548. DOI: 10.1016/j.cell.2018.12.015
  • [2] Koyama, S. et al. (2026). Exome-wide association study of blood lipids in 1,158,017 individuals from diverse populations. Nature Genetics, 58(6), 1268-1279. DOI: 10.1038/s41588-026-02613-y. Source paper: s41588-026-02613-y.pdf