Scientists from Columbia University and the Chan Zuckerberg Initiative developed GREmLN, a graph-based artificial intelligence model that better captures complex gene regulatory interactions than previous sequential language models. Trained on millions of single-cell gene expression profiles, GREmLN predicts how genes coordinate cellular states and responses, outperforming state-of-the-art models in cell type annotation and network understanding tasks. The approach addresses gene-gene interactions across entire genomes and supports accelerated training and computational efficiency. GREmLN advances computational biology by aligning AI architectures with biological networks, improving disease modeling and therapeutic target identification.