Application of Libra to my dataset

[RobertWhitener]

Hi,

I am interested in using Libra to explore different DE methods for my labs scRNAseq projects. However, I have some questions regarding exactly how to do it.

This is a human organ atlas project, and so we do not have "technical replicates" in our datasets. Instead, we have a large number of plate-based Smart-Seq2 libraries comprising 29 (and growing) individual organ donors spread across two different "Projects" which we are integrating and clustering using Seurat. The Seurat team currently advises that cell clustering be performed on the "integration" assay of the object (containing the "anchors"), but any DE analysis be performed directly on the "RNA" assay (normalized expression matrix).

As it is an atlas project, we have numerous types of cells in the libraries. We handle this by doing an initial data integration followed by manual cell type annotation to identify the major cell types, and then generate new Seurat objects containing 1 each of the major cell types.

From there, we are able to sub-cluster the cells, greatly expanding the total number of cell types we have.

We have generally been using the MAST algorithm to calculate DE genes across our sub-clustered cells. However, we often end up with large numbers of DE genes, and are interested in trying pseudo-bulk methods as well.

The read.me is written in a way (as I interpret it) that assumes there is a large number of technical replicates, rather than our large number of biological replicates.

Any possibility to help me understand the best way to approach using Libra for our type of data?

Thanks!
Robert

[jordansquair]

Hi Robert - we are referring to biological replicates not technical replicates. So in this case you would just need to set your replicate column to that which identifies the donor.

[RobertWhitener]

Hi Jordan,

Thanks! That is what I've been doing, so good to know I was on the right track. In this case, as we have two "projects" that we are integrating, I'm setting the label_col option to our "Project" column.

Best
Robert