search

neurogenomics / rare_disease_celltyping

Code, data and results associated with the "Rare diseases cell-typing" project.
6 stars 0 forks source link

Add some graphs showing the range of ontology terms associated with some cell types #25

Closed NathanSkene closed 1 year ago

NathanSkene commented 2 years ago

E.g. at the moment there are the plots showing that cardiomyocytes are most assocaited with cardiovascular terms

image

Jai also tinkered with these plots for breaking down what the 'Other' terms are... for this example it nicely highlighted that they still make good sense

image

bobGSmith commented 2 years ago

He used a sort of keyword search for that one if I remember correctly. I could get the disease descriptions from the HPO API and add that to the searchable info maybe? Or I could do it by branch, e.g. what percentage of significant immune cell enrichments are from the "Abnormality of the immune system" branch? @NathanSkene @bschilder

bobGSmith commented 2 years ago

I have been experimenting with some stuff. I'm just defining the enrichments based on what main HPO branch they come from, but could try a keyword thing like Jai. Rather than picking an individual cell type, I used all immune system-related cells or all cardiovascular system-related cells. Then counted the number of significant enrichments from each branch for that group of cells.

I think this could be quite a cool way to show more than one class of cell in a single figure: immune_cardio_radial Rather than just targeting those branches i picked out, i could just include all main HPO branches. Also could try make a function for it so can try any group of cells and any sub-branch i guess.

I started looking at bar and pie charts too (for immune only) immune_expected_bar

immune_expected_pie

@NathanSkene @bschilder

bschilder commented 2 years ago

@ovrhuman i really like that spider/radial plot! i think this is a great way to summarise these multi-dimensional findings in an intuitive way.

Rather than just targeting those branches i picked out, i could just include all main HPO branches. Also could try make a function for it so can try any group of cells and any sub-branch i guess.

I was thinking the same thing, users could choose which terms (at any level) they want to populate the "spokes" with. Could even be specific diseases.

He used a sort of keyword search for that one if I remember correctly. I could get the disease descriptions from the HPO API and add that to the searchable info maybe? Or I could do it by branch, e.g. what percentage of significant immune cell enrichments are from the "Abnormality of the immune system" branch?

Yes, i think breaking it down by branch makes sense to me. I'll leave it up to you to determine the exact level of resolution.

bobGSmith commented 2 years ago

@bschilder Yeah ill stick with branch descendants rather than keyword search stuff then. They are all just different options for subsetting the data but using the branches is a lot easier to work with