Enable selection from gene-gene x/y plot, and interactive linking with the main UMAP plot

[ambrosejcarr]

from @ademorree:

What would be really cool is if the software could replicate the cell gating strategies from FACS software. In FACS software software you can plot all cells in 2d for two parameters as we can also do in cellxgene; then in that graph you can lasso a population (say cells high for both parameters); then plot only those cells onto a next graph with 2 new dimensions. At all times you can see where in the original graph the selected cells exist. Of course, FACS deals with at most 20 parameters, so much less complex than our data. But a simple question I often face when looking at an XY plot in cellxgene is where are these cells in the underlying umap, do the cluster together or not, etc.

[baohongz]

We developed cellxgene VIP (Visualization in Plugin) to generate violin, stacked violin, stacked bar, heatmap, volcano, embedding, dot, track, density, 2D density, sankey and dual-gene plot in high-resolution SVG/PNG format. Dual-gene plot allows you to look at an XY plot to investigate where these cells are in the underlying umap based on expression level of gene A and B (Fig. 1b).

[vals]

Hi,

This issues relates to a couple of things I am interested in. Let me illustrate with a data set I have been working on.

If you find some clusters, it is good to look at QC values and see how they relate to your tSNE. Here is an example tSNE: Now, if I make a scatter plot of a couple of QC measures such as total count and MT fraction, I can see this: Currently you can only do scatter plots of gene expression in cellxgene, it would be very useful to be able to do scatter plots between arbitrary quantitative column in the obs field.

It would be nice to be able to lasso select the dense blob on the right. If it was linked with the tSNE, I would then be able to see that the dense blob corresponds to the larger population of cells, while the small clusters correspond to cells with low count and particularly high or low MT fraction:

Note also log scales everywhere, related to issue #1211

Going back to the pairwise genes idea for FACS style gates. Say I want to gate the cells as proliferating or not. I can use a pair of proliferation markers and gate the cells based on these:

It would also be interesting to access and plot columns from the obsm field. The obsm field is good for storing other useful values, for example for a subset of genes. Then a gate could be done using e.g. posterior expression frequencies from scVI: