Immune deconvolution updated heatmap

[sjspielman]

Purpose/implementation Section

From Issue #1230, this PR uses ComplexHeatmap to visualize immune deconvolution scores, with a matrix of cell types across samples (not groupings). The plot is therefore highly similar to that made for GSVA scores.

Directions for reviewers. Tell potential reviewers what kind of feedback you are soliciting.

Which areas should receive a particularly close look?

Figure is here

• Do we want to use this plotting strategy for two plots?
  • Some pros to this strategy -
    • Uses the same legend as other panels
    • By plotting samples, we avoid this situation where there were highly distinct scores within histologies but across molecular subtypes. Not summarizing by histology group means we aren't pooling distinct molecular subtypes
    • We may be able to add another panel(s) to "zoom in" on some molecular subtypes or similar?
  • Some cons -
    • Repetitive panels in the same figure (not sure this is a real problem)
• The color palette
  • Scores should be "normalized" within cell type groupings (here, rows), but there is no obvious procedure for doing this. Currently, I have implemented z-scores. The problem is the color palette. Z-score distributions are highly skewed with this five number summary - -1.902852956 -0.375620332 -0.139271522 0.002413132 47.912638453. We can't have a color ramp from [-2, 48]....

Reproducibility Checklist

• [x] The dependencies required to run the code in this pull request have been added to the project Dockerfile.
• [x] This analysis has been added to continuous integration.

Documentation Checklist

• [ ] This analysis module has a README and it is up to date.
• [ ] This analysis is recorded in the table in analyses/README.md and the entry is up to date.
• [ ] The analytical code is documented and contains comments.

[sjspielman]

Noting this failed because of CI time out.

Per discussions with @jaclyn-taroni, we may not want two heathap panels in figure 4. An alternative option that will still have the benefit of showing samples directly could be as follows -

• A faceted grid of cell types that show high amounts of variation across cancer groups
• X-axis is molecular subtypes, Y-axis is fraction (explore with(out) z-score normalization)
• Jitter or (or box/violin with jitter overlay) geom(s)

[sjspielman]

To explore this other plotting option, I have added this figure which shows T cell CD8+ and overall immune scores for EPN and HGG cancer groups, with molecular subtypes highlighted.

This plot does not seem very informative to me, except potentially to look at overall immune scores that xCell estimates. Feedback or thoughts on how to move this approach forward (or not?) would be great!

[jaclyn-taroni]

I'd facet by cancer_group as well. But I think we should go about selecting what gets displayed a different way. My knee jerk reaction would be to subset to features with higher variance but it seems like we might expect scores to be on a different scale?

[sjspielman]

I did make a grand viz to get a sense of overall variability within cell types (attached and aptly titled). There are a handful of cell types without much variance, but most have quite a bit of variance across some, but not all, subtypes. The molecular subtypes are proving tricky to integrate here, especially since sample size across them varies so widely. This is just colored by cancer groups, to get an overall sense of the variance in cell type scores. woah.pdf

Woops, entered too soon. My idea was to use this figure to pull out cell types of interest, but I don't see many striking differences across the subtypes so choosing is tricky. It seems driven mostly by sample size.

[jharenza]

Hi @sjspielman! Coming back to this. Now that #1243 is merged, can you recreate the tmp and "woah" figures with those cell types + the XCell immune score?

Also, will you remove any subtypes which are "To be classified" and add MB subtypes to the EPN/HGG tmp figure? I can take a look at the patterning then and see if it makes sense with the literature.

Thanks!

[sjspielman]

Yes, but..I think we should actually shut down this PR though and start with a fresh one focusing on xCell etc. The original heatmap I made for this PR was headed in the wrong direction and a cleaner PR will be easier for review I think. I'll link back to this discussion in a new one!

Actually quick clarification - are you referring to the boxplot as "tmp" (is this "temporary")? I might be missing an acronym here :)

[jharenza]

Oh yes, I abbreviated temporary as tmp here! Sorry, was writing it fast, and sounds good re new PR!