The abstract is read at the publisher; the summary is JClub's.
Fine-tuning Borzoi and AlphaGenome on HCHS/SOL personal genomes improves whole-blood gene expression prediction, but performance stays similar to variant-based linear models across Amerindian-, European-, and African-enriched groups.
- Why it matters: Reliable individual-level expression prediction in non-European, admixed populations is needed to interpret non-coding variants and understand regulatory mechanisms, yet newer sequence-to-function models are not well characterized there.
- What they did: The study fine-tuned Borzoi and AlphaGenome using whole-blood RNA-seq paired with personal whole-genome sequencing from HCHS/SOL, then compared them to variant-based linear models and assessed ancestry-enriched subgroup performance.
- The result: Model accuracy is similar across Amerindian-, European-, and African-enriched groups, with modest gains when training and evaluating within the same ancestry-enriched group; adding predicted epigenetic features improves prediction near promoters.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.