SpliceSelectNet: a hierarchical Transformer-based deep learning model for splice site prediction

SpliceSelectNet — hierarchical Transformer model for splice site prediction with a 100 kb contextual window, enabling detection of deep-intronic variants and long-range splicing effects

SpliceSelectNet is a hierarchical Transformer model trained on a 100 kb contextual window for nucleotide-resolution splice site prediction. Unlike existing predictors (SpliceAI, Pangolin) limited to a few kilobases, SpliceSelectNet captures distal regulatory effects and deep-intronic pseudo-exon-creating variants. In silico mutagenesis validates the functional importance of high-attention regions. The tool is open-source and surpasses the state of the art on published benchmarks.

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

Extending the context to 100 kb for splicing prediction is a major advance: deep-intronic pseudo-exon-creating variants are a frequent cause of genetic diseases unresolved by exome approaches. SpliceSelectNet is a significant update to the bioinformatics toolkit for splicing analysis in diagnostic genomics.