spatially variable genes (svgs)
abspangler13 opened this issue · 5 comments
-
ask tony about his code
-
re-watch Lukas's talk from 6/30. https://jhu-genomics.slack.com/archives/CR9NYA0BF/p1656599274806249
-
use this to solved curvature of gyrus question. Find spatially variable genes of the just the two clusters around the curvature. Hypothesis is that this phenomenon is drive by a gene expressed in a gradient.
-
meet w/ Lukas?
-
do nnSVG for k = 16 all pairwise comparisons. only works for one sample at a time.
-
run on specific pairs https://jhu-genomics.slack.com/archives/C01EA7VDJNT/p1658867952886359
-
figure out why it's taking so much memory. Or submit jobs to multiple queues.
-
do just these pairs https://jhu-genomics.slack.com/archives/C01EA7VDJNT/p1659115769482389?thread_ts=1658429571.622209&cid=C01EA7VDJNT
Pairwise: 5 vs. 9 (Layer 4?)
Pairwise: 4 vs. 16 (Layer 5?)
Pairwise: 7 vs. 13 (Layer "6A" vs. "6B")
Pairwise: 12 vs. 13 (layer 6 apex vs Layer 6B
Pairwise: 12 vs. 7 (layer 6 apex vs. layer 6A)
Pairwise: 16 vs. 12 (layer 5 apex vs. layer 6 apex) -
combine pairs into one new cluster label, run nnSVG with new cluster label as covariate. Doing this for the first 3 pairs all together first since there are no overlapping cluster labels. The other ones will have to be done separately.
Last round of this is still running. The output for this is confusing and bit unorganized so, feel free to contact me about what is what. It takes a long time to run so you probably don't want to have to do that again.
Ok, thanks! I'll take a look next week and maybe talk about it with @lmweber too.
@lcolladotor my last run of nnSVG failed. I tried to log into JHPCE to look at the log files, but I think my access is gone. Just putting this here for your information.
Thanks Abby!
See https://jhu-genomics.slack.com/archives/C01EA7VDJNT/p1665090349917619 for more.
I think that we can get nnSVG to run by avoiding the for loop with an array job + reduce memory loads.