Normalization of RNA-Seq Counts with ERCC spike-ins

Question

[Berta]

I have RNA-Seq data of an experiment that involves global shift in transcript abundance and were spike-ins were added (to try to overcome bias based on this global shift). Reads were aligned and counted using STAR and RSEM. I have been trying to find consistent approaches to normalise RNA-Seq counts using ERCC Spike-ins counts, but all I find are quite disperse approaches.

One of the approaches I read consisted in normalising the RPKMs of their genes by dividing them by the sum of the RPKMs of their spike-ins. Influenced by this, I aimed to normalise TPMs the same way: TPM of genes divided by the Sum of the TPMs of spike-ins.

Even though it seems quite straightforward to me, I do not find anyone doing something similar and I am quite new to the field. Would this spike-in-normalised TPMs be reliable? If not, can someone point out where the mistake would be?

[GenoMax]

Not something you will like but here goes:
https://support.bioconductor.org/p/88413/
What is ERCC spike-in really saying about my sequencing run?

Answer 1

[Trivas]

You could check out the RUV-seq package, see here, specifically RUVg. This will give you the estimated factors of unwanted variation that you can include in your DESeq2 design formula as well as the normalized counts (note: do NOT use these for differential expression).