Learning Gene Association Networks Using Single-Cell RNA-Seq Data: A Graphical Model Approach

on YouTube ↗34m

" summary

Framework learns gene association graphs by fitting node-conditional regressions with a zero-inflated negative binomial (covering NB and Poisson limits), testing edge presence with deviance tests inside a PC algorithm search, and proving consistency plus family-wise error control under no latent confounders. Simulations show the surprising result that for low UMI counts a simple Poisson often suffices (even with simulated 90% zero inflation), while ZINB matters for high-count regimes. Method reconstructs networks on 20,000+ olfactory single cells, recovers ChIP-seq neighbors, and ships as an R package on GitHub.

AI-written summary · about

" chapters (14)

1. 0:00Intro
2. 1:09The evolution of gene expression measurements
3. 2:28Data generating distribution for RNA-seq
4. 3:10Overdispersion
5. 3:39The negative binomial distribution
6. 4:47Enters single-cell RNA-seq
7. 6:45Are genes independent?
8. 13:55Applications to gene expression
9. 17:41PC algorithm
10. 19:59Simulation results (large mean)
11. 20:50Simulation results (small mean)
12. 21:24Zero inflation: a real issue?
13. 24:01Transcription factor dynamics across development
14. 26:40My research group