Filtering genes after TPM normalization

Question

[JACKY]

Filtering genes is one of the steps that I do in all my analysis. I usually filter the counts RNA-seq data according to a specific threshold. In my current work, I have no access to the counts data, I only get the normalized data. All data is normalized with TPM approach.

I'm wondering if it is still a good idea to filter genes even though the data is normalized? if so, what would be a good approach for that? since almost every package in R filters counts data, and not TPM data.

Thank you!

Answer 1

[Meisam]

While not a fan of it, I’ve seen papers where authors used a cpm or tpm value cutoff of 1 to filter their count table. Since this threshold would be anyhow arbitrary, you might want to use your best guess for what roughly the total reads per sample could be. Like if you know they used a MiSeq machine and you have three samples you would expect to have around 3-5 million reads per sample before normalization.