how to process the RNA-seq data? log10(count+1) or log2(count+1) or TPM or FPKM?

[wang99999shang]

when i use the rawcount of RNA-seq data, the result is bad. Should i use log10(count+1) or log2(count+1) or TPM or FPKM? Which one is better? Thank you!

[Chen-Guanming]

I guess RNA-seq data should be normalized by Seurat::NormalizeData(), becuase I found the Scissor just got the scRNA-seq normalized data and bound with bulk data by checking the Scissor code.

common <- intersect(rownames(bulk_dataset), rownames(sc_dataset))
sc_exprs <- as.matrix(sc_dataset@assays$RNA@data)
dataset0 <- cbind(bulk_dataset[common,], sc_exprs[common,])

[jzheng25]

I have similar questions. I think default Seurat using log normalization with 10000 scale factor. Does it mean we should re-normalize fkpm or tpm x/sum(x)*10000 and then take log1p? But normalize.quantile seems do normlization by ranking. That comes to my question whether the independent variable(expression) to be normally distribution to fit the assumption of underlying regression?

[JZHT-jiangzhou]

i think we can use the LogNormalize function to normalize the bulk RNA seq, the Seurat_preprocessing function shows : normalization.method = "LogNormalize", scale.factor = 10000