Can compaRe's similarity function be used with single cell RNA sequencing data?

[morchalabi]

[morchalabi]

• Potentially yes. What makes, more importantly, single cell cytometry data and single cell sequencing data (scRNAseq or snRNAseq) different is their inherent amount of variance. In scRNAseq, owing to the sparsity of data (e.g. UMI count matrix), there is a much higher amount of variance. This means that unlike cytometry data that all markers (e.g. membrane or cytoplasmic markers) can be used for similarity measurement, in scRNAseq, you cannot use all markers (genes), but first should follow the standard preprocessing steps to reduce the variance. Additionally, this excessive amount of variance induces nontrivial signal shift aka batch effect. Therefore, for being able to compare 2 samples, they should be integrated (batch effect removal). The standard steps could be:

  • for each sample:
    • removing poor-quality cells (dead cells, multiplets and empty droplets) from the count matrix,
    • library-size normalization of the count matrix,
    • subsetting the normalized matrix with the most variable genes,
  • integrating the two matrices,
  • Z-transformation of the integrated matrix and
  • applying PCA on the z-transformed matrix

The resultant PCA matrix (after separating each sample) can be used as input to compaRe.