Use with CyTOF data?

[ashipde]

Can we use BERMUDA for mass cytometry (CyTOF) data? Unlike scRNA seq. data, CyTOF data may have only 20-40 variables (markers).

[txWang]

Hi,

The intuition of BERMUDA, which is using domain adaptation to align similar cell clusters across different batches, should work on CyTOF data. However, the design of the whole BERMUDA pipeline is for scRNAseq data and we only tested the method on scRNAseq data in our paper. There are several potential issues I could think of when trying to apply BERMUDA to CyTOF data.

1. We used Seurat to identify cell clusters and select highly variable genes. I think with 20-40 markers, you don't necessarily need to identify highly variable genes. However, I haven't tried Seurat on cell cluster identification using CyTOF data.
2. We used MetaNeighbor to identify similar clusters across different batches. I am not sure if their method will work on CyTOF data.
3. Since you are working with a lot fewer variables, you might want to adjust the default network in BERMUDA to a smaller network, but the design of losses should work fine.