Math and Biological Data - Jingyuan Hu | Ep 17
Jingyuan Hu — Postdoctoral Researcher, Harvard Biostatistics
Jingyuan Hu on inferring where a cell is heading from one snapshot of single-cell data, why more cells of the same kind hit a ceiling for biology's AI models, and learning the mathematics on YouTube.
Key Takeaways
A snapshot holds the dynamics
Sequencing freezes each cell at one moment, yet where cells are heading can be inferred from it, like cars in a photograph of a highway. 6:42
FlowMap infers the landscape
His method learns Waddington's landscape from the molecular profiles, to ask why two similar stem cells choose different fates. 13:00
Validation is the bottleneck
Single-cell data makes hypotheses easy to generate, but checking them can mean looking at one gene at a time. 16:08
More cells hit a hard ceiling
Random matrix theory shows a saturation point, past which more cells of the same kind add very little biological information. 25:01
Expertise can be built on YouTube
With little mathematics in his formal training, he learned it course by course on YouTube, starting with Gilbert Strang's linear algebra. 32:08
“If you are curious, you can build expertise outside your formal training.”
Chapters
- 0:00Intro2 min
- 1:32What single-cell data can show2 min
- 3:57Manifolds: the geometry of cell data1 min
- 5:20Dynamics from a single snapshot6 min
- 11:00FlowMap and a cell's fate4 min
- 15:13Validation, the field's bottleneck4 min
- 19:01Does biology just need more data?3 min
- 22:11Random matrix theory in simple words2 min
- 24:33The hard ceiling on more cells2 min
- 26:22Is AI for biology overhyped?3 min
- 29:02The virtual cell and treatment2 min
- 30:55Learning math from YouTube4 min
- 35:17Physics models vs. black-box AI6 min
- 41:09Creating new biology6 min
Mentioned in This Episode
About this Guest
Jingyuan Hu
Postdoctoral Researcher, Harvard Biostatistics
Computational biologist at the Harvard T.H. Chan School of Public Health, where he builds statistical and AI models for single-cell data, among them FlowMap. He also teaches the mathematics behind them on his YouTube channel.