using covariates for integrating 10x Visium slides

[sdwien]

Dear Harmony Team, I recently started to work with 10x Visium data, and one of the projects comprises 8 samples analysed on 2 Visium slides. The sample set has several variables which I annotated in the Seurat objects as follows:

orig.ident	mouse.ident	batch.ident	biol.ident
sample1	mouse1	batch1	control
sample2	mouse1	batch1	control
sample3	mouse2	batch2	control
sample4	mouse3	batch2	control
sample5	mouse4	batch1	treated
sample6	mouse4	batch1	treated
sample7	mouse5	batch2	treated
sample8	mouse6	batch2	treated

The 2 control and the 2 treated samples from batch1 are technical replicates = slices from the same tissue block each. The samples from both batches contain similar numbers of spots (~ "cells"), but batch1 contains ~ 50% more reads per spot than batch2. When just using Seurat::merge, the technical replicates overlap completely in a UMAP12 plot, but do not merge with the other biological group from the same batch and also not with the samples of the same biological group in the second batch.

Following the workflow for batch correction described on the 10x Genomics, I tried RunHarmony using single covariates or combinations thereof, with varying results. Using only "batch" as covariate, the 2 (resp. 4) samples from batch1 still cluster apart from the others.

I would appreciate your suggestions regarding the following: Would you recommend to use all of the variables for Harmony? In general, would you use the biological state as a covariate at all? What would be the use of running Harmony (if any) using only the orig.ident as covariate, which in my case is the individual sample labels basically? I have not played with different theta values, do you think this would be necessary?

Many thanks for this useful tool, and for your help. Best regards, Sophia