A novel AI model named GREmLN (Gene Regulatory Embedding-based Large Neural model) has been developed by Columbia University and Chan Zuckerberg Initiative researchers to improve predictions of gene regulation and cell state transitions. GREmLN integrates graph-based architectures informed by gene regulatory networks to capture complex, long-range gene interactions not addressed by prior language model-based approaches. The model was benchmarked against state-of-the-art single-cell RNA models, demonstrating superior gene expression reconstruction and cell type classification capabilities across healthy and diseased human cells. GREmLN is part of CZI's ongoing efforts to build virtual cell simulations that enhance understanding of cellular behavior and response to interventions. The approach promises advances in computational biology and treatment prediction.