search

MarioniLab / miloR

R package implementation of Milo for testing for differential abundance in KNN graphs
https://bioconductor.org/packages/release/bioc/html/miloR.html
GNU General Public License v3.0
318 stars 20 forks source link

Mass cytometry data #343

Open NicoFuenza opened 6 days ago

NicoFuenza commented 6 days ago

Hi!

I have been experimenting using Milo to analyse mass cytometry data. The data I try to feed into the Milo pipeline has been clustered using FlowSOM, and I am using UMAP for 2D projection. My SCE file contains therefore UMAP coordinates (for 10 million cells), but no PCA coordinates. I have tried running the pipeline using the UMAP coordinates when reduced dims are needed, and it works fine until I try to run testNhoods (with glmm); it ends up running for hours without any output (even with only 10 000 cells).

I have also tried to calculate PCA coordinates (without recalculating the UMAP coordinates), and running the pipeline. Now I get some results, however, with adjusted p-values equal to 0.99. Would fewer neighbourhoods (larger k and smaller p) affect the adjusted p-values?

Thank you!

MikeDMorgan commented 5 days ago

Hi @NicoFuenza - I would strongly discourage using the UMAP co-ordinates to build a graph as they can distort the distances between individual cells. Could you post some example code and outputs/plots to get a better idea of what you are dealing with concretely.