Closed lcolladotor closed 1 year ago
After meeting with @stephaniehicks @kmartinow and others, we'll go more conservative and instead of dropping Sp02D01
, we'll do the k28 domains that are WM annotated.
https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position_noWM/ is now updated after this change https://github.com/LieberInstitute/spatialDLPFC/commit/d56af05d8249f4ef1b8402f6e331e9ff4b83f885. I'll post a summary later tonight.
Ok, the new summary for the analysis without white matter (this time, without SP2806, 16, 17, 20 and 28 instead of Sp02D01) is:
SP28
ones)spatialLIBD
app: https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position_noWM/Here are the updated screenshots from the top 25 (FDR < 5%) genes for the enrichment tests:
Notably, in posterior > rest
, MBP is the 7th (versus the 6th before) and MOBP is the 13th (same rank) as before when we were dropping Sp02D01. cc @kmartinow @stephaniehicks
Though note that the gene set enrichment that we were seeing earlier with SFARI is gone now at https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position_noWM/. Also, the number of genes decreased a bit in this version of the no white matter analysis (vs dropping Sp02D01).
I didn't take a screenshot of the previous version of the no white matter analysis for the gene set enrichment. Here's the one from with white matter https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position/
So right now, I'm still inclined to trust the with white matter results more than the without white matter ones. cc @lahuuki @kmaynard12. This is likely due to some WM genes showing up in some layers / spatial domains as we get closer to WM, and without having a clear WM cluster (Sp09D06) https://github.com/LieberInstitute/spatialDLPFC/blob/c8fbcbf051a13def3f57f2a4201e00ca6ac1e8da/code/analysis/16_position_differential_expression_noWM/logs/01_create_pseudobulk_data_noWM.9.txt#L333 (in this version Sp09D09 is very rare https://github.com/LieberInstitute/spatialDLPFC/blob/c8fbcbf051a13def3f57f2a4201e00ca6ac1e8da/code/analysis/16_position_differential_expression_noWM/logs/01_create_pseudobulk_data_noWM.9.txt#L336) the without white matter analysis suffers (more) at trying to avoid WM vs GM genes.
Anyways, that's one way of seeing it. There are likely other ones.
If @kmaynard12 (and/or @kmartinow) like one version of this analysis, we can move on to some ideas @lahuuki and I were discussing today. These include (I'm likely missing some other ideas):
Sp02D01 > Sp02D02
(proxy for WM > GM
; could be Sp28D06 + 16 + 17 + 20 + 28 vs rest; aka the 0-1 indicator variable @stephaniehicks mentioned today) in the x-axis vs anterior > test
, plus 2 more facets for middle > rest
& posterior > rest
. The idea that some genes with high position
enrichment t-stats but low WM > GM t-stats (top left corner & top middle, but not top right part of the scatter plot) might be of more interest.position
.@kmaynard12 let me know if you have questions about these results.
I need to make the supplementary figure describing these results after what we discussed in a recent "spatial meeting".
Completed with plots at https://github.com/LieberInstitute/spatialDLPFC/tree/main/plots/09_position_differential_expression, mainly https://github.com/LieberInstitute/spatialDLPFC/blob/main/plots/09_position_differential_expression/density_histogram_enriched_tstats.pdf and https://github.com/LieberInstitute/spatialDLPFC/blob/main/plots/09_position_differential_expression/sce_pseudo_gene_explanatory_vars_k09_large.pdf
Now that #124 is done, we'll need your help @kmaynard12 checking the
position
enriched genes for either or for both the analyses with and without dropping the white matter spots (Sp02D01
spots). That is, dropping 14,353 spots (https://github.com/LieberInstitute/spatialDLPFC/blob/5a7a09f032660de72d751063796ef688c62dab6f/code/analysis/16_position_differential_expression_noWM/logs/01_create_pseudobulk_data_noWM.9.txt#L109). In these analyses we are adjusting for the spatial domain, age, and sex. While I did the analysis for k from 3 to 28, the results below are only for k = 9 (Sp09). Note that Sp09, Sp09D06 gets completely wiped out https://github.com/LieberInstitute/spatialDLPFC/blob/5a7a09f032660de72d751063796ef688c62dab6f/code/analysis/16_position_differential_expression_noWM/logs/01_create_pseudobulk_data_noWM.9.txt#L331.To summarize a bit what we have now, here's a little list:
spatialLIBD
apps where the difference is on thelayer-level
portion of the app. The links are at https://github.com/LieberInstitute/spatialDLPFC#interactive-websites.anterior > rest
but alsoanterior > middle
for example. The one for the top 25 genes (FDR < 5%) is small enough that GitHub shows a preview, the other one you'll need to download to your computer to open it.Here are the links to these apps and files:
With white matter
spatialLIBD
app: https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position/Without white matter spots (aka
Sp02D01
ones)spatialLIBD
app: https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position_noWM/Example
As an example, the 2nd ranked gene for
anterior > rest
without WM (aka https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position_noWM/) is GPM6A https://www.genecards.org/cgi-bin/carddisp.pl?gene=GPM6A&keywords=GPM6A that has the following info:Similarly, it's also the 2nd ranked gene for
anterior > rest
at https://libd.shinyapps.io/spatialDLPFC_Visium_Sp09_position/Though BASP1 https://www.genecards.org/cgi-bin/carddisp.pl?gene=BASP1&keywords=BASP1 goes from 9th ranked with WM to 3rd ranked without WM.
I suspect that if we did a Venn diagram of the FDR < 5% genes for a particular statistical test (say
anterior > rest
), the overlap will be quite high between both analyses (with and without WM) since we are adjusting for the spatial domains while computing these statistics. We could also do scatterplots of the t-statistics. But well, for now, I think that you wanted to check some of these results first before we make any plans for new plots.Screenshots top 25 enriched genes
With WM
Without WM
(cc @kmartinow)