The latest advancement in AI technology by the Chan Zuckerberg Initiative (CZI) is set to revolutionize how researchers comprehend the behavior of cells. The GREmLN (Gene Regulatory Embedding-based Large Neural model) is a groundbreaking model that focuses on the intricate networks that govern cell behavior, making it easier to tackle complex biological issues like cancer.
Published on the pre-print server bioRxiv, GREmLN is designed to provide insights into how genes within a cell cooperate and how abnormalities in these interactions can lead to diseases such as cancer or neurodegeneration. This innovative approach, according to Andrea Califano, the president of Chan Zuckerberg Biohub New York, offers a biological perspective to leverage AI for uncovering new insights into health and disease.
Unlike traditional AI models, GREmLN delves into the “molecular logic” of gene interactions within cells, akin to a conversation happening inside the cell. By capturing these critical changes, the model allows scientists to pinpoint the earliest signs of disease and potential targets for novel treatments.
Trained on a vast dataset from Chan Zuckerberg CellxGene, GREmLN will soon be expanded to address pressing biological and medical queries. This expansion aims to enable researchers to detect the initial stages of cell malignancy or neuronal damage before irreversible changes occur.
Theofanis Karaletsos, senior director of AI at CZI, emphasizes that understanding cellular behavior entails comprehending the network of conversations within each cell. GREmLN excels at capturing this complexity like never before, paving the way for predictive systems that simulate cell behavior.
In the future, GREmLN could play a pivotal role in various research applications, from enhancing the efficacy of cancer therapies to safeguarding brain cells from inflammation-induced damage. Moreover, the model could assist in forecasting cellular responses to new drugs, thereby boosting their success rates in clinical trials.
Part of CZI’s virtual cell platform, GREmLN can be accessed by researchers globally, alongside a quick start tutorial and the model’s codebase on GitHub. Together with other biomodels like TranscriptFormer, GREmLN contributes to the organization’s mission of advancing biomedical research, disease diagnosis, and therapeutic development.
In conclusion, GREmLN signifies a significant leap in AI tools designed to aid scientists in deciphering the intricate logic of cellular processes. By bridging the gap between AI and biology, CZI’s innovative models are poised to make groundbreaking contributions to the field of life sciences.
For more information:
Mingxuan Zhang et al, GREmLN: A Cellular Regulatory Network-Aware Transcriptomics Foundation Model, bioRxiv (2025). DOI: 10.1101/2025.07.03.663009
This article is provided by the Chan Zuckerberg Initiative.
