New AI Method Predicts Gene Activity in Diverse Human Cells

A new artificial intelligence system reliably predicts gene activity in human cells, offering researchers unprecedented insights and potential advances in tackling complex diseases.

• Researchers at Columbia University created an AI model capable of forecasting gene expression in previously unstudied cells
• The system provides new ways to investigate disease mechanisms, including a form of paediatric leukaemia
• Findings hint at the possibility of exploring “dark” regions of the genome linked to cancers and other conditions

The quest to understand the complexities of human cells has taken a significant step forward with the development of a novel artificial intelligence (AI) method that accurately predicts which genes are active in almost any cell type. Researchers at Columbia University Vagelos College of Physicians and Surgeons have published their findings in Nature, suggesting that biology may be on the verge of a new predictive era.

“Predictive generalizable computational models allow to uncover biological processes in a fast and accurate way. These methods can effectively conduct large-scale computational experiments, boosting and guiding traditional experimental approaches,” says Raul Rabadan, professor of systems biology and senior author of the new paper.

This effort comes amid growing interest in AI-driven approaches to biological questions. In 2024, the Nobel Prize in Chemistry was awarded to researchers for their use of AI to predict protein structures—a discovery that brought AI’s potential in molecular biology to the global spotlight. However, predicting the behaviour of genes inside an intact human cell has remained a far more complex challenge.

Building a Predictive Biology

Many traditional methods in biology focus on describing how cells work or measuring reactions to specific disruptions, such as disease mutations. They rarely provide a lens into how cells might behave under varying or novel conditions. According to Rabadan, if scientists can accurately forecast a cell’s activities, this could “transform our understanding of fundamental biological processes” and turn biology “from a science that describes seemingly random processes into one that can predict the underlying systems that govern cell behavior.”

In this new study, researchers collected gene expression data from over 1.3 million healthy human cells. They combined information on the genome sequence and the accessibility of specific genetic regions, training their AI model to identify the “grammar” of cell regulation. “Previous models have been trained on data in particular cell types, usually cancer cell lines or something else that has little resemblance to normal cells,” Rabadan says. By focusing on normal cells from a wide range of tissues, the team created a system able to generalise to unseen cell types.

“Here it’s exactly the same thing: we learn the grammar in many different cellular states, and then we go into a particular condition—it can be a diseased or it can be a normal cell type—and we can try to see how well we predict patterns from this information,” explains Rabadan. The group tested the approach on cell types never encountered by the model, and it successfully matched real-world experimental data.

Implications for Cancer and Beyond

To illustrate how predictive AI might push new research frontiers, the investigators used their method to examine an inherited form of paediatric leukaemia. “These kids inherit a gene that is mutated, and it was unclear exactly what it is these mutations are doing,” says Rabadan, who also co-directs the cancer genomics and epigenomics research programme at Columbia’s Herbert Irving Comprehensive Cancer Center.

By analysing the cells, the system revealed that these mutations disrupted the interaction between two essential transcription factors—information subsequently verified through laboratory work. This not only aids understanding of how certain childhood cancers develop but also highlights the broader potential of the model to illuminate hidden biology.

Moving beyond protein-coding genes, Rabadan’s team expects that exploring the genome’s “dark matter” could be the next frontier. “The vast majority of mutations found in cancer patients are in so-called dark regions of the genome. These mutations do not affect the function of a protein and have remained mostly unexplored. says Rabadan. ‘The idea is that using these models, we can look at mutations and illuminate that part of the genome.’”

With this method, scientists now have an opportunity to study uncharted genomic regions and develop a clearer understanding of how unknown or poorly characterised mutations might fuel cancer or other diseases. The work represents a noteworthy stride towards a predictive biology that may ultimately offer new avenues for targeted therapies, as well as fundamental insights into the shared language of all human cells.