I noticed that many scRNA-seq papers normalize TPMs to 10k or 100k as opposed to 1M (as the abbreviation defines them). It doesn't really matter since you are just moving the decimal point, so why mess with an established convention?

Additionally, the values might be log-transformed. In that case, log(x+1) would affect the data less if the total is 1M.

  • 4
    $\begingroup$ There tend to be far fewer counts in single cell experiments. By reducing the scaling factor you are avoiding having massive expression values for your samples. $\endgroup$ – GWW Dec 14 '18 at 19:52
  • $\begingroup$ Could you provide some links, so that we can check if there is some explanation in the methods section? $\endgroup$ – llrs Dec 15 '18 at 10:06
  • $\begingroup$ @GWW Isn't "massive" subjective? With standard TPM, the highest possible value is 1M. It doesn't seem that big. Arguably, by switching to 10k, your small values become too small (extra 0s at the beginning). $\endgroup$ – burger Dec 17 '18 at 17:06
  • $\begingroup$ @llrs I haven't seen an explanation yet. Here is an example paper from last week (reputable journal and lab): doi.org/10.1016/j.celrep.2018.11.056 . "... We also excluded outlier cells from the MPPs group which were likely caused by cell contamination. To correct for batch issues, the two experimental batches were centered to have the same mean log(TP100K+1) (hereafter referred to as log(TPM), where TPM stands for ‘transcripts per million’) per gene. PCA was done in R with scaled..." $\endgroup$ – burger Dec 17 '18 at 17:08
  • 1
    $\begingroup$ @burger: It isn't uncommon to scale single cell experiments by the median cell library size. That way small values typically won't be too small and large values won't be too large. People also choose 10,000 because that's roughly around the library size for a single cell. $\endgroup$ – GWW Dec 17 '18 at 17:54

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.