Future thought: Explore disagreements between references

This plot seems to me to be possibly the most useful one here. I am particularly interested by just comparing the blueprint to HPCA data, which seem to actually disagree on the labels "common myeloid progenitor" and "granulocyte monocyte progenitor" for a significant portion of cells. Blueprint "wins" here, but I honestly don't know what to make of that: is it a better reference, or just better matched to this dataset?

I think it might be interesting to plot a matrix of comparison between the the combined data and each reference, ideally some kind of annotation of the relationships among labels: when two references disagree, is there a parent/child relationship between the labels? We'd have to dive further into the ontology to create such a plot, so I think it is out of scope for this notebook. We would want to spend some time sketching out designs before trying to actually generate the plot, so this is a long-term thought.

_Originally posted by @jashapiro in https://github.com/AlexsLemonade/sc-data-integration/pull/219#discussion_r1207092879_

Begin by reviewing the existing notebook and the plot comparing Blueprint with HPCA data. The proposed work is to sketch and then generate a matrix comparing combined data with each reference, including ontology relationships such as parent and child labels; done means an agreed design and a useful comparison plot.

Written by the indexing model from the issue text.