Dear all,

I would like to ask you please for a suggestion : we do have a set of multiple scRNAseq experiments CTRL and STIM (in the same tissue) at 1, 2, 3, 4 months i.e:

CTRL1month, CTRL2months, CTRL3months, CTRL4months

STIM1month, STIM2months, STIM3months, STIM4months

and I have done the analysis by comparing :

a) CTRL1month vs STIM1month

b) CTRL2months vs STIM2months

however, for each comparison, the CLUSTER NUMBERS differ i.e. :

the CLUSTER_NUMBERS of the comparison :

a) CTRL1month vs STIM1month,

differ from the comparison :

b) CTRL2months vs STIM2months ...

is there any way we can keep the same CLUSTER NUMBERS over MULTIPLE COMPARISONS ?

in SEURAT, we've looked at the CONSERVED MARKERS between CLUSTERS in CTRL1 vs STIM1, and CTRL2 vs STIM2; the problem is that some of the clusters share too many conserved markers ...

thanks a lot,

-- bogdan

You won't be able to force the clustering algorithm to give you the same cluster number for the "same cell type" across batches. I mean, the clustering algorithm working on one batch of data doesn't even know about the other batch, so how can it possibly keep the numbers consistent? And what is a cell type anyway?

Just figure out which clusters match up to each other. There are multiple ways to do this:

• batchelor (in the BioC-devel version, 1.3.9) has a clusterMNN() function that can be used to do mutual nearest neighbor matching on clusters. You can extract across-batch groups of related clusters from the metadata() of the output.
• SingleR has a matchReferences function, just feed it two datasets with some sort of annotation and watch it construct a matrix of mutual assignment probabilities. You can visualize this on a heatmap to easily figure out which clusters match between batches.

And there's no need for all-caps, I can read perfectly fine without them.