At our virtual Showcase Symposium 2020, Allen Institute for Brain Science teams present project talks highlighting their work. Please use this forum thread to ask questions for the speakers of the Patch-seq Pipeline and Discoveries talk: Staci Sorensen, Rusty Mann, Kim Smith, @Fahimeh, @racheld, and @gouwens. Please see the full list of Showcase 2020 forum threads to ask questions about other talks.
@Fahimeh, what distribution model are you using for your Kullback-Leibler estimates? Or is the probability coming from histograms?
In your patch-seq procedure, when you extract nucleus, did you also take some cytosol too? If yes, will that confound interpretation if you compare gene expression between cells?
We do also extract cytosol. It’s not a confound, though, as (1) we do care about genes whose mRNA is outside the nucleus and (2) we are benchmarking the cells to a dissociated single-cell that also includes cytosol. We’re not trying to isolate the nucleus alone; it’s more that we found that transcriptomic data quality improved when we pulled nucleated patches.
Hi, Thanks for the question. Well, in order to see how good a patchseq cell is, I compare its probability of mapping to the mapping probability of all the FACS cells. For that I do the following: 1- Map all the FACS cell onto FACS cell, 2- For the FACS cell, I have the clustering results as well, from Tasic 2018 paper. 3- Then For each FACS cell type, I can define a mapping probability to all other cell types. 4- Now I map Patchseq cells on to FACS cells, 5- Then I compare their probability of mapping to the FACS probability of mapping using KL divergence. If they are good quality cells, their divergence from FACS going to be small and they will be highly consistent cells. Otherwise they will be among inconsistent or moderately consistent cells.