Mapping high level histology color labels -- Part 1

[cansavvy]

Purpose/implementation Section

What scientific question is your analysis addressing?

This PR addresses Step 1 of 4 discussed in #897. This PR adds that main notebook described where we attempt to group biospecimens into "high level" histology groups that will be more practical for plotting purposes.

Note that subsequent PRs will deal with the updating of the README and retirement of histology_color_palette.tsv in favor for the mapping table, palettes/histology_label_color_table.tsv, that is created in this notebook.

What was your approach?

Copied from the notebook added in this PR itself:

# Purpose: 

The histology label variables included in `pbta-histologies.tsv` from data releases are not always useful for visualizing the full set of biospecimens due to the large number of different values.
Having too many different possible values makes the colors harder to distinguish.
In addition, there are some groups that are represented by only a very few samples; giving such groups a distinct color may be counterproductive.

The goal of this notebook is to use the currently existing `broad_histology` groups from `pbta-histologies.tsv`, to form 10-15 "high level histology" group labels that can used for plotting purposes.

## The output table

The output of this notebook is a TSV file: `palettes/histology_label_color_table.tsv` that contains the following fields:

**Copied from `pbta-histologies.tsv`**:    
- `Kids_First_Biospecimen_ID` (from `pbta-histologies.tsv`)  
- All the original histology label variables (`broad_histology`, `short_histology`, etc.)  
  
**Created in this notebook**:  
- `display_group` - the high-level histology labels that should be used for plotting   
- `hex_codes` the direct colors that should be used for plotting  

With this info, `histology_label_color_table.tsv` can be used by all plots and figures that summarize high level data  while displaying histology information. 

# How `display_group` is made:

Here's how `broad-histology` groups are [combined into the higher-level groupings of `display_group`](#declare-new-equivalent-groups).

1) "Neuronal and mixed neuronal-glial tumor" and "Diffuse astrocytic and oligodendroglial tumor" were labeled as LGAT in previous releases so we will group these into `Low-grade astrocytic tumor` in `display_group`. 

2)  `Germ cell tumor`  samples are combined into the `Non-CNS tumor` group. 

3) `Benign tumor` and `Non-tumor` biospecimens are combined into a `Benign` group. 

4) Anything not in the above categories gets its `broad_histology` label carried over. 

What GitHub issue does your pull request address?

#897

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

Which areas should receive a particularly close look?

• How do we feel about this notebook and its ability to deal with upcoming releases' versions of pbta-histologies.tsv
• How do we feel about these broad histology groupings? Aka how scientifically wrong are they? (Note that we've always known "high level" groupings will not be more accurate or specific; that is not the goal here, the goal is to be broad and have something easier to visualize in a plot.

Is the analysis in a mature enough form that the resulting figure(s) and/or table(s) are ready for review?

Yes. The ones included can be.

Results

What is your summary of the results?

We have a list of colors and 16 + Normal groups. I have not tried it on an official plot yet. (upcoming PR)

Reproducibility Checklist

• The dependencies required to run the code in this pull request have been added to the project Dockerfile. -- no new packages are needed.
• This analysis has been added to continuous integration. -- I've added it at the very top because the first few CI tests will need it once Make mapping histology groups for plotting -- Part 2 implementing the use of the mapping table #898 is addressed and those plots use the table created here.

Documentation Checklist

• This analysis module has a README and it is up to date. -- No, this will be in a subsequent PR. The README will require quite a bit of updating.
• This analysis is recorded in the table in analyses/README.md and the entry is up to date. -- it is not because its not located in the analyses folder.
• The analytical code is documented and contains comments.

[jashapiro]

This looks like a good start. I think my main comment is that we should try to be a bit more intentional about grouping things before resorting to "Other". There may well be reasonable groupings that take some of the small-count histologies and link them to large-count histologies that they are related to rather than just to other small-count groups. It may be helpful to have a table which shows the integrated_diagnosis values for each broad_histology to assist with such grouping.

My other suggestions are mostly about display. This is a notebook for internal evaluation, so having more easily visible data at each stage in tables is more valuable than some of the summaries or figures.

I also suggest treating "Other" as a special group, with a defined color that remains constant.

With regards to your question:

For ease of downstream use, I'm wondering if we should drop everything that's not Kids_First_Biospecimen_ID summary_group and hex_codes, because downstream we'll need to drop everything.

I think we should keep as much information as possible for now. It is potentially easier to evaluate groupings using the full table, and the cost of keeping it is essentially zero.

[jashapiro]

I would here again print out the full table to make it easier to evaluate what has happened.

[jashapiro]

One thing when making the colors is that I feel like we should treat "Other" as special, and always assign it a standard color, something neutral & grey like #A0A0A0 (though I am not sure how that fits with colorblind-friendly schemes).

[cansavvy]

Alternatively, we could drop this "Other" group from the large summary figures.

[jharenza]

We consider treating "Benign" as the grey group @jashapiro and @cansavvy - because it's more like a control and we would be less interested (maybe) in those genomics.

[cansavvy]

@jharenza , we could use some more domain knowledge about these histologies at this point.
In particular, which of these display_groups (which are mostly from broad_histology) would be most reasonable to combine into other groups?

display_group	n
Lymphoma	2
Melanocytic tumor	2
Other tumor	2
Other astrocytic tumor	6
Metastatic tumors	9
Tumor of pineal region	10
Histiocytic tumor	12
Choroid plexus tumor	21
Pre-cancerous lesion	27
Non-CNS tumor	29
Mesenchymal non-meningothelial tumor	48
Meningioma	59
Tumors of sellar region	73
Benign	78
Tumor of cranial and paraspinal nerves	83
Ependymal tumor	174
Embryonal tumor	339
NA	833
Low-grade astrocytic tumor and adjacent	1034

[jaclyn-taroni]

Without looking at the code, I wanted to point out that the NA should be normal samples. So you may want to include some kind of filtering to tumor samples if you have not already.

[cansavvy]

Without looking at the code, I wanted to point out that the NA should be normal samples. So you may want to include some kind of filtering to tumor samples if you have not already.

I did have that previously but with the changes I made most recently with Josh's review we lost that. I'll add it back.

[cansavvy]

@jharenza this is where we set up the equivalent groups which are then recoded in the next section.

[jharenza]

Hi @cansavvy -

1. "Neuronal and mixed neuronal-glial tumor" and "Diffuse astrocytic and oligodendroglial tumor" were labeled as LGAT in previous releases so we will group these into Low-grade astrocytic tumor in display_group.

This is not true - These are three separate entities and really should not be combined. The neuronal and mixed glial tumors comprise of many distinct tumors, so they had a wider array of short_histology. The Diffuse astrocytic and oligodendroglial tumors were all HGAT (High-grade astrocytic tumors) and should be kept as a separate group. So, I would keep those three groups, especially since they have a bulk of the tumors.

In particular, which of these display_groups (which are mostly from broad_histology) would be most reasonable to combine into other groups?

I think we can combine:

Lymphoma	2
Melanocytic tumor	2
Other tumor	2
Metastatic tumors	9
Non-CNS tumor	29

to Non-CNS or other tumor

I think you can combine Other astrocytic tumor into LGAT for now, since it is a low-grade astrocytoma. Possibly, we can update that broad_histology in v19.

I think that will leave us with 15.

[cansavvy]

@jharenza , thanks for these comments. I'm going to implement them and then re-request you for a review when its ready.

[cansavvy]

I think that will leave us with 15.

This brings us to 16 + Normal. Are there any other combos that seem reasonable or should we leave it at that?

[cansavvy]

I would be more prone to dropping Benign tumor first, then Other if we want to drop more.

If we are talking about dropping that completely from the figures, than I think we'd probably keep this data in the table for now, but change that at the figure scripts?

But I was more talking about groups we could combine into other groups.

[jharenza]

Ah, yeah I think that is as far as we can combine right now.

[jharenza]

Hi @cansavvy! Looks good- made a few minor comments, but otherwise it looks good!

[jharenza]

We consider treating "Benign" as the grey group @jashapiro and @cansavvy - because it's more like a control and we would be less interested (maybe) in those genomics.

[jashapiro]

We consider treating "Benign" as the grey group @jashapiro and @cansavvy - because it's more like a control and we would be less interested (maybe) in those genomics.

This makes sense to me. Maybe make both Benign and Other "special" so that their colors aren't random?

[jharenza]

Sure!

[jashapiro]

Looks good. Just a question about what to do with "Normal" colorwise, and some concern about capitalization consistency between files.

[jashapiro]

One typo, but looks good!