Compare HC Annotations with Azimuth Results

lahuuki commented 2 years ago

[x] Upload and run on Azimuth Human Motor cortex app
[ ] Compare with current HC annotations, use to help annotate Excit layers

lahuuki commented 2 years ago

Use Azimuth annotations as "guide" to split and further annotate clusters (not merge as we previously thought). Recreate this heatmap map with annotations for: number of 1vALL marker genes (specifically groups with none), best layer + max layer correlation value. Maybe annotate Azimuth cell types by our broad classifications.

co-authored w/ @lcolladotor

lcolladotor commented 2 years ago

We are updating this one to:

Summary heatmap

~~- [ ] Update number of marker genes results from https://github.com/LieberInstitute/DLPFC_snRNAseq/issues/7#issuecomment-1232064775 on the y-axis. Although maybe we'll drop this: let's talk about it.~~ ~~- [ ] Update with number of nuclei by hc on the y-axis~~ ~~- [ ] Update with number of nuclei by Azimuth on the x-axis~~

[ ] Replace max_cor and max_cor_layer by layer_label from https://github.com/LieberInstitute/DLPFC_snRNAseq/blob/main/processed-data/05_explore_sce/spatial_registration_cor_details_hc.csv on the y-axis (label as NA the ones with asterisks, so poor under layer_confidence)
[ ] Replace max_cor and max_cor_layer by layer_label from https://github.com/LieberInstitute/DLPFC_snRNAseq/blob/main/processed-data/05_explore_sce/spatial_registration_cor_details_azimuth.csv on the x-axis (label as NA the ones with asterisks, so poor under layer_confidence)
[ ] (keep) broad cell type on both axes + confirm with Azimuth website/Kristen

Preliminary clusters

https://github.com/LieberInstitute/DLPFC_snRNAseq/blob/main/plots/05_explore_sce/06_explore_azimuth_annotations/azimuth_v_prelim-Excit_02.png and related ones. We'll put them together into a multi-panel supplementary figure.

[ ] Actually, I think it might be best to have them as a single heatmap so the columns are all in the same order, but labeling on the y-axis which hc cluster they come from.

While we won't split the hc clusters based on this, we can mention these observations in case others want to split the hc clusters later.

Known markes across Azimuth

Violin plots of known markers. I think this is what's already at https://github.com/LieberInstitute/DLPFC_snRNAseq/blob/main/plots/05_explore_sce/06_explore_azimuth_annotations/Azimuth_basic_mathys_markers.pdf. But maybe there's another similar PDF.

[ ] Extract some highlights to show why the results from Azimuth have some issues themselves. We'll also mention on the text that L1 and L4 are missing in motor cortex.

lcolladotor commented 2 years ago

For the number of marker genes, I know that you weren't too enthused with how pheatmap binarizes the continues metric into a color range with 4 values. With that in mind, maybe ComplexHeatmap would be more useful where we can add a barplot on the side like at https://jokergoo.github.io/ComplexHeatmap-reference/book/heatmap-annotations.html

lcolladotor commented 1 year ago

@lahuuki I think we can close this one, right?

LieberInstitute / DLPFC_snRNAseq