Cells in a developing organism constantly communicate with each other, exchanging chemical signals to determine their fate – to become a liver cell, a brain cell, or something else. Understanding this complex correspondence has long been difficult: it was believed that each signal worked in its own way in each cell type, requiring painstaking study of each pathway individually.
A new AI tool, IRIS, developed by scientists at MIT, breaks this barrier. It has discovered that each signaling pathway leaves a unique "fingerprint" – a characteristic pattern of gene activity. Importantly, these fingerprints have proven to be universal for the same pathway, regardless of cell type. This means that the history of signals received by a cell can now be reconstructed much faster and more efficiently.
IRIS, being a neural network, analyzes the overall gene activity of a cell and predicts which signaling pathways were active at specific moments of development. The model is trained on a vast dataset of the response of thousands of embryonic stem cells to various combinations of signals. This allows for a deeper understanding of the fundamental mechanisms of cell communication and provides a practical tool for controlling stem cell development in the laboratory.
Scientists have already used IRIS to predict the signals necessary for lung cell development. Experiments have confirmed the model's predictions, paving the way for more reliable creation of laboratory models of lung tissue for studying diseases such as asthma or cancer, and for testing new treatment methods.