1086-ChromBERT: A Foundation Model for Regulatory Networks

Researchers have developed ChromBERT, a genomic foundation model designed to decode the complex "syntax" of interactions among over 1,000 transcription regulators. By pre-training on a massive dataset of human ChIP-seq profiles, the model creates interpretable representations of regulatory networks that function across diverse cell types. ChromBERT excels at cistrome imputation, accurately predicting binding events in unseen cells by using fine-tuning prompts from chromatin accessibility or transcriptomes. Beyond simple prediction, the model identifies key regulatory drivers behind cell state transitions and enhancer activity without requiring new, expensive experiments. Ultimately, this technology overcomes the challenge of sparse data to reveal the hierarchical mechanisms governing gene expression in both healthy and diseased tissues.

References:

• Yu Z, Yang D, Chen Q, et al. ChromBERT: A foundation model for learning interpretable representations for context-specific transcriptional regulatory networks[J]. Cell Genomics, 2026.