Best way to perform single-cell RNA-seq normalization

Question

[igor]

Is there an optimal way to do single-cell RNA-seq counts normalization? Most RNA-seq normalization tools are designed with bulk samples in mind and assume a certain distribution of reads. With bulk RNA-seq, you have millions of reads per sample with relatively consistent coverage across samples. With single-cell experiments, you have much fewer total reads and many low expressing genes will be completely absent from many samples. I am curious how much I should worry about that and if there is a good way to account for that.

Answer 1

[Devon Ryan]

For single-cell experiments people tend to normalize the spike-ins and use the resulting size factors. I would be very hesitant to use the actual gene/transcript/whatever counts for normalization in cases like.

[Sandeep]

Any particular tool we can implement the same? I understand DESeq uses the size actor for normalization. Does that mean DESeq is a good method to do it?

[Devon Ryan]

Yup, DESeq2 would work fine for this. There are some single-cell RNAseq oriented packages out there, such as SAMstrt. I don't work with single-cell data at the moment, so I can't say how good/bad things like SAMstrt are, but you should have a look at them.

[Sandeep]

Will do. Thanks.

[igor]

I am not sure about the spike-in controls. For example, Fluidigm (leading commercial solution) doesn't use spike-ins with the standard protocol.

[Devon Ryan]

Interesting, that makes life difficult. We use MARS-seq internally, cheaper and scales to many more cells.