Decoding and Translating the Languages of Biology: towards de novo cell design
- Date published: 2026-10-17
- Speakers: Jian Shu (guest talk to the Levin lab)
- Format: Guest talk
- Watch: https://www.youtube.com/watch?v=SiTlGdMKQNo
- Topics: AI and biology
In one paragraph
Jian Shu presents tools for measuring and redesigning cell states, aiming at designing new cells from scratch. He starts from iPS-cell reprogramming, moves to methods that use optimal transport on single-cell data to trace how cells change over time, and then to methods that predict gene expression from label-free images (Raman to RNA). He closes by calling for a shared, organized dataset of cell images and measurements so AI can suggest the next experiment.
Main ideas (plain words)
- His PhD work found stem cells can be made by balancing rival lineage factors (a "seesaw" model), not only with the classic stem-cell factors.
- Waddington-OT uses optimal transport to link single-cell snapshots across days and reconstruct cell trajectories.
- Raman to RNA predicts single-cell gene expression from Raman imaging, a label-free way to look at living cells. Lipids turned out enriched in senescent cells.
- He proposes a community resource like ImageNet or the Protein Data Bank, but for cells.
Key quotes (word for word from YouTube's auto-generated captions)
"how do we actually engineer cells, which is the fundamental units of life."
https://www.youtube.com/watch?v=SiTlGdMKQNo&t=30s"maybe we can build this image omics net could be like the image net and protein"
https://www.youtube.com/watch?v=SiTlGdMKQNo&t=2190s"Maybe with all of this, and then this could be our future dream that we can achieve. That's the novel cell design."
https://www.youtube.com/watch?v=SiTlGdMKQNo&t=2250sSource: YouTube auto-generated captions, speakers not labeled; wording is provisional. Timestamps point to the start of the 30-second stretch that holds the quote.