[I first posted this here but realize that this forum is more suitable for this question]

Hello,

I am currently working with a single cell dataset from a frog colon and would like to estimate the origin of the cells (i.e. the region of the colon). There exists a dataset that has performed bulk RNAseq of the 8 different colon sections.

My idea is to use this dataset to predict the origin of my single cells. The only issue is, that I only have access to 8 samples (each corresponding to one colon section). My question is, if it is still valid to use SingleR in that case, or if the sample size of the reference is too small. I noticed that in the vignette the size of your reference dataset was much bigger (>700).

Any help is much appreciated!

The vignette data are "large" because it is single-cell data and each cell is considered in this case a "replicate". With bulk RNA-seq you need biological replication to do any reliable stats and in turn derive any reliable signatures for scoring. I doubt that with n=1 per section that is possible. Hence, SingleR might perform poorly simply because the reference is poor.