ERCC - should they be treated separately?

[jtheorell]

Hi again,
I am now having a new think about ERCC. Have you tested what happens with spike-ins with any of the datasets for the paper? I am thinking that it might be more natural to treat them separately, i.e. as they are non-biological, they cannot be supposed to follow the same distribution as the rest. On the other side, they need to be transformed too to be possible to use down-stream, and as they are so few in number compared to the genes, they might not have a major impact, even if they are on average much more highly expressed than the average normal transcript. So what would you recommend? For now, I will create qUMI counts for the ERCC with all the other transcripts included, and then exclude them to calculate the qUMIs for the true transcripts.
Best regards
Jakob

[willtownes]

Good question! Honestly I have no idea how to deal with ERCC. I feel like it's worthy of a whole research paper to investigate their statistical properties and how to best normalize. For now I would just say quminorm as a method is only verified on non-ERCC count data. If it works with ERCC that would be great but I can't make any promises.

[jtheorell]

Very good. Then I will stick to my approach above for now, as I need to have the ERCC on the same scale to be able to use them (and as my previous tests showed that the correlation between the known input and the calculated concentration of the ERCC strengthened with the QUMI transformation).