Cutoffs for TPM values

[mht]

Hi,

I'm using RSEM to compute gene expression values for two different RNA-seq sample. RSEM outputs tau value which I then converted to TPM value by multiplying it by 10^6.

1) Is there a minimum cutoff value I should set for TPM value? I see some papers filter out genes with TPM < 1 or TPM < 5 before proceeding to further analysis. How do I determine what cutoff should be used?

2) In a set of gene expression results, is it possible to determine which genes are considered highly expressed and which are expressed at low levels? I've see some papers use 1 < TPM < 9 as low and TPM >=100 as moderate/highly expressed, while others use percentiles (e.g. >= 75% percentile as highly expressed).

Thanks in advance.

[CPCantalapiedra]

We seem to have the same issue than several years ago.
We have a set of RSEM files from different samples and our results from EdgeR point out that we have a lot of noise since we have not filtered out the transcripts based depth of coverage or counts (TPMs, ...).

We have seen that, in our data, a TPM of about 1 use to have around 3x DoC. However, we are not sure how to establish a cutoff or if it is enough by obtaining such relationship and choose the DoC we are confident with.

In addition, we have the question whether we should reanalyze edgeR results after filtering RSEM files, or if "translating" the filtering to the edgeR results would suffice.

Any hint?