Variant-resolved prediction of context-specific isoform variation with a graph-based attention model.

Area of Science:

• Genomics and Molecular Biology
• Computational Biology and Bioinformatics

Background:

• Eukaryotic genes generate multiple transcript isoforms, crucial for transcriptome and proteome diversity and functional regulation.
• Genetic variations can alter RNA processing signals, affecting isoform structure and abundance, but modeling these effects at full-length isoform resolution is complex.

Purpose of the Study:

• To introduce Otari, an attention-based graph neural network framework for predicting tissue-specific differential isoform abundance.
• To enable isoform-resolved interpretation of genetic variant effects by integrating sequence-derived signals.

Main Methods:

• Otari was trained on human genomic sequences and long-read transcriptomes across diverse human tissues and brain regions.
• The framework integrates sequence-derived epigenetic and post-transcriptional signals to predict isoform abundance.
• Otari was applied to large-scale variant datasets, including an autism cohort.

Main Results:

• Otari successfully predicts tissue-specific differential isoform abundance.
• The model uncovers isoform dysregulation patterns not detectable at the gene level.
• Variant-driven changes in isoform abundance and microexon usage implicated in autism pathophysiology were identified.

Conclusions:

• Otari provides a powerful resource for large-scale, isoform-level analyses across multiple tissues.
• The framework enhances the interpretation of genetic variant effects on transcript isoforms.
• Otari facilitates a deeper understanding of disease mechanisms, such as autism, at the isoform resolution.