RegFM
RegFM is a context-aware foundation model for human transcriptional regulation.
It treats regulation as a dialogue between cis-regulatory sequences (CREs) and trans-acting regulators (transcription factors and chromatin regulators), coupling long-range CRE representations with TF/CR activity.
Trained on large-scale ENCODE and CELLxGENE transcriptomic profiles, RegFM learns gene-centered regulatory representations that generalize across unseen cellular contexts.
Model description
- Inputs: long-range cis-DNA sequence features + cell-context TF/CR and expression signals
- Outputs: gene expression predictions and regulatory representations usable for downstream tasks
- Framework: PyTorch
Intended uses
- Gene expression prediction in unseen cellular contexts
- Cis-regulatory element annotation
- Bivalent promoter / dosage-sensitivity related analyses
- Perturbation-response prediction
- Interpretable analysis of cis–trans regulatory interactions
Code & demo
Code, training/prediction scripts, and a PBMC leave-one-out demo (predict on held-out CD8 TEM 1) live on GitHub:
https://github.com/ZjGaothu/RegFM
pip install git+https://github.com/ZjGaothu/RegFM.git
# or clone and: pip install -e .
Citation
If you use RegFM, please cite:
Zijing Gao, et al. RegFM: an interpretable context-aware foundation model for human transcriptional regulation. bioRxiv, (2026).
(DOI will be added upon public release.)
Contact
gzj21@mails.tsinghua.edu.cn
License
MIT