Comparing scores across runs

[liezeltamon]

Hi @ivokwee,

For my dataset, I have pseudobulked samples per cell type producing separate sample x gene tables per cell type. The number of genes after filtering lowly expressed genes are different across these tables/cell types, which could also influence the gene set score comparison across cell types.

I have run plaid per cell type to make overall workflow faster and to be able to batch correct before plaid without missingness due to some rarer cell types not being in a sample. This was fine for within cell type group comparisons, but I do want now to compare gene set scores across cell types, and I think I would have to rerun plaid() just to make sure the median normalisation accounts for all (celltype-samples) right?

Thank you for the help!

Best,
Liezel

[ivokwee]

In Plaid, normalization happens after calculating the gene set scores. So if you want, you can just median (or quantile) normalize the gene set scores yourself after plaid, and not do the recalculation if you don't want. But I guess plaid is so fast that you also just can run plaid again.

Again, if your counts are not normalized before running plaid, then normalization of the gene sets score (either by plaid internally or manually) is needed to account for different total read counts per sample/celltype.

Also, be careful to use at least 100 gene sets or more, because for smaller number of genesets the median normalization of gene set score will not work well.

BR. Ivo.,