Researchers at the Broad Institute of MIT and Harvard, in collaboration with ETH Zurich and the Paul Scherrer Institute (PSI), have developed a new AI-driven computational framework that helps biologists understand the complete state of a cell by intelligently integrating data from multiple experimental measurement types. Traditionally, scientists study cells using different methods — from gene expression to protein levels to morphology — and then manually piece together what’s going on, which is slow and often incomplete. The new system learns not only what data are shared across these modalities, but also which parts come uniquely from each measurement, giving a more holistic view of cellular behaviour.
This AI framework treats multimodal cellular data like overlapping sets in a Venn diagram, separating information that many techniques capture from details that only one measurement provides. By identifying where information came from, researchers can better interpret complex cellular processes — such as how cells respond to stress, communicate signals, or undergo disease-related changes — without running countless individual experiments. In tests on synthetic and real single-cell datasets, the method successfully distinguished shared and modality-specific information, an advance that could accelerate discoveries in cell biology and disease research.
One practical application of this approach was pinpointing which measurement modality best detected a specific protein linked to DNA damage in cancer patients — helping scientists decide what tools to use in future experiments. This work could improve understanding of disease mechanisms and even assist in planning treatments for conditions such as cancer, neurodegenerative disorders (like Alzheimer’s), and metabolic diseases such as diabetes, where complex cellular changes play a crucial role.
The researchers say their method could also help answer fundamental questions about which cellular measurements provide the most valuable insights and when additional experiments are actually needed. By automating this integration of cell data, AI enables faster, more accurate interpretation of biological systems and could become a standard tool for scientists seeking deeper insights into how cells function in health and disease.
