A unified framework for quantitative splicing and transcript prediction.

Deep learning trained end-to-end on splice site, splice junction and transcript usage across multiple mammalian species

SpliceAI2 is a deep learning model trained end-to-end on splice site, splice junction and transcript usage across multiple mammalian species. It reconstructs complete transcripts directly from genomic sequence and considerably improves the identification of splice-altering variants and the quantification of their effects across benchmarks spanning cryptic splicing, splicing quantitative trait loci and massively parallel reporter assays. Human population datasets independently validate the predicted effects, through evidence of natural selection and associated changes in plasma protein abundance. In a rare disease cohort, predicted splice-altering variants are highly enriched in disease-associated genes and account for 15% of the excess genetic burden, with approximately half residing in deep intronic regions likely missed by exome sequencing.

Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.

Preprint, not peer reviewed. Predicting not only that a variant disrupts splicing but also which transcripts result is exactly what is needed to argue an RNA effect before any functional test. The clinically relevant figure is the share of deep introns in the excess burden, which argues for systematically analysing these regions on genome data; the abstract, however, gives neither the cohort size, nor a diagnostic gain, nor model availability.

Analysis by Dr Thibaut Benquey