Cryptic splice site activation by rare GT/AG-creating variants in Genomics England blood RNA-seq (SpliceAI2 paper, auPRC)

Stable ID: rna-splicing-20261013-spliceai2-protocol-gel-gtag-detection

areas

dna-genomes; rna-transcriptomes

contexts

clinical_research

protocol

Same labelling pipeline as the GTEx GT/AG benchmark applied to Genomics England samples (novel junction only in carriers, from personalised diploid-genome alignments); auPRC of each model's score at the created site.

version

bioRxiv v1 Fig. 4g

metric

auprc

metric direction

higher

source locator

Main 'Contribution to rare genetic disease' paragraph 4; Methods 'Variant effect benchmarks' (GTEx and GEL GT/AG); Fig. 4g

limitations

Blood RNA-seq only: splicing changes in genes not expressed in blood are not observable, so nulls can include variants that splice in other tissues.; The GTEx rule that a cryptic gain is seen in more than one sample applies to GTEx only; the paper does not state a replacement rule for Genomics England.; Only overall auPRC is printed; no stratum or uncertainty values.; The rule-based motif-scoring baseline used for this use case (for example MaxEntScan) is not evaluated.; Developer-reported: the SpliceAI2 authors (Illumina) built the benchmark, chose comparator settings and ran every model; AlphaGenome was scored by co-authors at the University of Oxford.; Training overlap: SpliceAI2 was trained on all human genes except those on chromosomes 21 and 22, and the benchmark is genome-wide, so most variants lie in genes seen in training. Supplementary Table 1 does not list Genomics England data among the training sources.; The paper does not say whether essential splice variants were excluded from the Genomics England benchmark; the exclusion is stated for the GTEx and OpenSplice benchmarks (Fig. 1d).; No confidence intervals or resampling uncertainty are printed.; Preprint (bioRxiv v1), not peer reviewed.

• uses data: Genomics England rare-disease participants with WGS and blood RNA-seq (Jaganathan et al. 2026)
• assessment: AlphaGenome on Genomics England GT/AG cryptic splice activation (auPRC, SpliceAI2 paper)
• assessment: AlphaGenome raw splice site score on Genomics England GT/AG (auPRC, SpliceAI2 paper)
• assessment: Pangolin on Genomics England GT/AG cryptic splice activation (auPRC, SpliceAI2 paper)
• assessment: SpliceAI on Genomics England GT/AG cryptic splice activation (auPRC, SpliceAI2 paper)
• assessment: SpliceAI2 on Genomics England GT/AG cryptic splice activation (auPRC, SpliceAI2 paper)
• assessed by: Select a patient-RNA splicing-variant validation workflow