IRIS uses RNA to infer which developmental signals a cell received and helped refine a cell culture protocol
IRIS uses RNA to infer which developmental signals a cell received and helped refine a cell culture protocol
On September 8, Nature Methods published a paper on IRIS. The model was trained on human stem cells exposed to known combinations of six developmental signals, then applied to single-cell data from mouse embryos.
During embryonic development, neighboring cells send chemical signals to one another. The combinations and sequence of these signals influence what a cell will become. Computational methods typically look for a signaling molecule and its receptor, which can indicate that communication between cells is possible. An RNA profile records which genes are active in a cell at a given moment, providing a snapshot of its state. An evaluation of 400 models using cell atlases showed that simply expanding these collections of RNA profiles soon stops improving model performance. IRIS is first trained on cells exposed to known combinations of signals. It then searches RNA profiles for the signature of a signaling pathway, the chain of events through which a cell responds to an external signal.
To teach the model to recognize these signatures, the researchers cultured human embryonic stem cells and exposed them to known combinations of signals at different stages. They then measured RNA in individual cells under each condition. IRIS linked each controlled exposure to the signature it left in the activity of many genes. When given data from mouse embryos, the model reconstructed the expected sequence of signal combinations in the cell lineages that give rise to the foregut and heart muscle.
The authors then tested a prediction for cells destined to become respiratory mesenchyme, the tissue surrounding the lung bud. IRIS estimated an elevated probability of WNT and BMP activity. In cultured foregut tissue from mouse embryos, WNT activation expanded the region expressing two respiratory mesenchyme marker genes, Tbx4 and Foxf1; a WNT inhibitor eliminated Tbx4 expression.
The authors used this finding to adjust a protocol for human stem cells. They began WNT stimulation on the fourth day of differentiation and extended its duration. TBX4 expression increased, while an intermediate level of WNT maintained FOXF1 expression. Inferring signaling activity from RNA thus helped identify when to introduce WNT to obtain more cells with respiratory mesenchyme characteristics.