Towards universal modeling of transcript isoform expression levels

A universal graph-based model can infer human transcript isoform expression levels across 29 tissue types using 324 IHEC samples, consistently outperforming tissue-specific training.

• Why it matters: Accurate isoform-level expression modeling from epigenetic features is needed to quantify gene regulation in normal and disease states, but prior work used immortalized lines and gene-level summaries.
• What they did: Using paired transcriptomic and epigenomic data from the International Human Epigenome Consortium, the study built graph-based models that combine location-specific epigenomic features with multiple gene-gene relationship types.
• The result: For each sample, models leveraging many other tissue types consistently beat models trained only on that sample, supporting the feasibility of a universal approach to isoform inference across tissues.

AI summary of the abstract · may contain errors · not medical advice · read the paper.