RegFM: an interpretable context-aware foundation model for human transcriptional regulation
Gao, Z.; Sun, Y.; Wang, H.; Jiang, R.; Liu, Q.
Show abstract
Transcriptional regulation is governed by interactions between cis-regulatory elements (CREs) and trans-acting regulators in a context-specific manner. Although DNA and single-cell foundation models have enabled modeling regulatory biology at scale, most represent either sequence or cellular state alone, limiting their ability to capture context-dependent gene regulation. Here we present RegFM, a context-aware foundation model for human transcriptional regulation. RegFM treats transcriptional regulation as a dialogue between cis-regulatory sequences (e.g., CREs) and trans-acting regulators (e.g., transcription factors (TFs) and chromatin regulators (CRs)) by coupling long-range CRE representations with TFs and CRs activity. Trained on large-scale ENCODE and CELLxGENE transcriptomic profiles, RegFM learns gene-centered regulatory representations that generalize across unseen cellular contexts. In a wide range of tasks, including gene expression prediction, cis-regulatory element annotation, bivalent promoter and dosage-sensitivity classification, and perturbation-response prediction, RegFM consistently improves over existing methods. RegFM emerges as a scalable and interpretable framework for modeling human transcriptional regulation and provides insights into context-dependent gene regulatory programs.
Matching journals
The top 6 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
- Multi-scale deep tensor factorization learns a latent representation of the human epigenome 96%
- EvoAug: improving generalization and interpretability of genomic deep neural networks with evolution-inspired data augmentations 95%
- Geometry aware graph attention networks to explain single-cell chromatin state and gene expression 95%
Similar papers in this journal
- Iterative improvement of deep learning models using synthetic regulatory genomics 94%
- Estimation of allele-specific fitness effects across human protein-coding sequences and implications for disease 94%
- High-resolution mapping of regulatory element interactions and genome architecture using ARC-C 94%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.