xDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction
- 1School of Biomedical Sciences, University of Hong Kong, Hong Kong, China
- 2School of Computing and Data Science, University of Hong Kong, Hong Kong, China
- 3School of Artificial Intelligence, Shenzhen University, Shenzhen, China
Abstract
Large-scale genomic language models (gLMs) hold promise for modeling gene regulation, yet their ability to capture personal gene expression variations remains unresolved. We developed xDecoder, a unified decoding framework that utilizes gLMs and sequence-to-function (S2F) embeddings to learn how personal genetic variation shapes gene expression from paired genome-transcriptome data. Compared to the pretrained genomic models, xDecoder with personalized DNA–RNA training makes cross-individual prediction tractable for seen genes in a few-shot setting. However, zero-shot prediction at unseen loci remains unreliable and gene-dependent, revealing a cross-locus transfer bottleneck of current sequence models. Experiments incorporating individual-level chromatin accessibility suggested that regulatory-state information important for unseen-locus prediction is not fully captured by current DNA-only models. Overall, these results highlight the potential utility of the few-shot setting, the limitations of DNA-only models, and point toward multi-omic, variant-aware frameworks as a promising direction for building personalized regulatory models.
Related
- An open structure-guided RNA foundation model for generalizable prediction
- A multispecies foundation model for genome function prediction, gene annotation and sequence design
- Cross-tissue graph attention networks for semi-supervised gene expression prediction
- CRISPR/Cas9-mediated UBC knockout impairs CD44 expression and reduces the metastatic potential of 23132/87 gastric cancer cells
- Genome-Wide Identification and Characterization of the UMAMIT Gene Family in Brassica napus: Evolution, Expression, and Potential Association with Oil Content.
BibTeX
@article{li2026xdecoder,
title = {{xDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction}},
author = {Shumin Li and Ruibang Luo and Yuanhua Huang},
journal = {Molecular Systems Biology},
year = {2026},
month = jul,
doi = {10.1038/s44320-026-00238-1},
url = {https://doi.org/10.1038/s44320-026-00238-1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - xDecoder unlocks the potential of genomic foundation models for few-shot personal gene expression prediction
AU - Shumin Li
AU - Ruibang Luo
AU - Yuanhua Huang
PY - 2026
DA - 2026/07/01
JO - Molecular Systems Biology
DO - 10.1038/s44320-026-00238-1
UR - https://doi.org/10.1038/s44320-026-00238-1
AB - Large-scale genomic language models (gLMs) hold promise for modeling gene regulation, yet their ability to capture personal gene expression variations remains unresolved. We developed xDecoder, a unified decoding framework that utilizes gLMs and sequence-to-function (S2F) embeddings to learn how personal genetic variation shapes gene expression from paired genome-transcriptome data. Compared to the pretrained genomic models, xDecoder with personalized DNA–RNA training makes cross-individual prediction tractable for seen genes in a few-shot setting. However, zero-shot prediction at unseen loci remains unreliable and gene-dependent, revealing a cross-locus transfer bottleneck of current sequence models. Experiments incorporating individual-level chromatin accessibility suggested that regulatory-state information important for unseen-locus prediction is not fully captured by current DNA-only models. Overall, these results highlight the potential utility of the few-shot setting, the limitations of DNA-only models, and point toward multi-omic, variant-aware frameworks as a promising direction for building personalized regulatory models.
SN - 1744-4292
ER -