A Foundation Model for Predicting Transcription Across Human Cells

Understanding transcriptional regulation is key to decoding cellular functions, but existing computational models struggle to generalize across diverse cell types. The new General Expression Transformer (GET) addresses this challenge by leveraging chromatin accessibility and sequence data to predict gene expression with experimental-level accuracy across 213 human fetal and adult cell types. GET outperforms existing models in identifying regulatory elements, transcription factor interactions, and long-range genomic regulation, even in unseen cell types. Notably, it uncovered distal regulatory regions in fetal erythroblasts and a lymphocyte-specific transcription factor interaction linked to leukemia risk. Its adaptability across sequencing platforms makes it a powerful universal tool for gene regulation research, providing new insights into cell-type-specific transcriptional networks.