Genome BiolJClub
Fine-tuning sequence-to-expression models on personal genome and transcriptome data.
Genome Biology · · Journal Article · Open access
Rastogi, Reddy + more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
Fine-tuning Enformer on paired personal genome and transcriptome data improves gene expression prediction on held-out individuals for genes seen during fine-tuning, matching variant-based linear models.
- Why it matters: Out-of-the-box sequence-to-expression models underperform at predicting expression variation across individuals from personal genome sequences, leaving a gap in individualized prediction accuracy.
- What they did: The study fine-tunes Enformer using multiple strategies on paired personal genome–transcriptome datasets and evaluates performance on held-out individuals and held-out genes.
- The result: Fine-tuning improves predictions for genes observed during training, including from held-out populations, but does not improve generalizability to held-out genes with unseen sequences and variants.
The findingWhy it mattersWhat they didThe result
- Open access
- 15 cites