Pulin Li

My lab is really interested in understanding how cells form patterns in a tissue, and why these patterns might be important for organ development and function. One of the major mechanisms of pattern formation is for the cells to talk to each other using signaling pathways. Such communication guides cells to decide what cell type to become, or where to go, to build tissue patterns and organs.

Identifying which signaling pathways are involved in each step is crucial for understanding tissue patterning, but often takes years of intense work. Instead, we asked if we can infer the signaling history of cells from their current states, which are often read out using single-cell RNA sequencing. Traditionally, people thought that when different cell types receive the same signal, they would have different transcriptional responses. We wondered whether there might be a hidden signature of the responses that is shared across cell types, like a fingerprint unique to that signal that we can see in cells after the signal has been activated.

We used machine learning to try to find or extrapolate such fingerprints, and to our surprise it worked really well. Now we can generate lots of interesting hypotheses about what combinations of signals are used at each step of development. This could be very useful for regenerative medicine that relies on engineering cells towards specific fates. This new approach for mapping how cells talk to each other could also have broad implications in other biological contexts, such as cancer and immunology.