Thread by @MartinBJensen

Very excited (enough to get up at 5am) about this Temporal Single Cell Analysis organized by @singlecellomics. First talk by the amazing Caroline Uhlers, recently snagged from MIT by @ETH_en #SCOGtempSC #SingleCell

Livetweets here. Apologies in advance to any sophisticated 'ML on scSeq/spatial' presenters whose work I misinterpret/misrepresent, still a novice to that field.

Very nice talk by Caroline. Two parts: 1) mapping RNAseq and images to the same latent space, to enable timecourse measurements (w images) of (inferred) RNA state. Seems like WIP but cool.

2) predicting cell responses. Her group tried to apply scGen (cool tool from @MohammadLotfol1) to the Broad CMAP but found it didn't work well. Turns out contractive elements of training set take over, unless you over-parameterize. Moderator @fabian_theis took this very well😄

@GioeleLaManno talks about (of course) #RNAVelocity. Brings up need for negative (randomized unspliced) control for latent space bias before applying to real data. Answering a lot of audience Qs about velocity. What does length of arrow mean in UMAPs? Does it work for #snucSeq?

More #RNAvelocity from @VolkerBergen. He's working on a model with dynamic learning with subpopulations rather than classic steady-state model. Seems to work better for cycling progenitors, and IDs genes likely contributing to velocity. Also nice cell fate collab w @dana_peer.

Now @dana_peer on about recent work by @SettyM, & work expanding on #RNAvelocity with CellRank: Markov modeling the propagation of velocities across #CellState manifold. This improves certainty about eventual fates when states overlap. Applying to lung cancer dedifferentiation.

Great job organizing and moderating @fabian_theis!
Nice inclusion of speaker breakout rooms with during coffee break. Closer to feeling like a live conference (Is a throng throng of human bodies or dozens of video tiles more overwhelming for speakers?🤔). Wetlab session up next.

Had to do some work and missed talked by Florian Erhard. Now @AlexandervanOu1 talks about ways to gain information about the distant past of RNAsequenced cells: Copy number variations (in tumor organoids), SNP-linked somatic mutations, lentivirus marking. Very cool ideas.

*Uhler not Uhlers

Final talk by Nikolaus Rajewsky, about metabolic labeling for timecourse data, looking at transcriptome in gene expression not latent space, and also about @LifeTimeIni. Latter is extremely cool and will be topic of another thread. Great job moderating & organizing @AE_Saliba.