TPM Cutoff For Venn Diagrams

Question

[jehammo]

Have a question concerning cutoff for Venn Diagram generation. Comparing microglia model systems for expression of microglia signature and microglia-enriched genes. Would like to observe three group Venn diagrams between model systems, ie iPSC-microglia, monocyte-derived microglia and primary adult microglia. My specific question is how to determine cutoff from TPM data for positive expression of a given gene? Seems arbitrary, but are there specific statistical or accepted methods to identify cutoff? For example, a TPM cutoff of 50 will provide about a 10-12% positive identification from 38k complete genomic genes. From there the subset of genes will be restricted to 882 microglia genes.Thanks in advance.

[igor]

This related thread may help: TPM values of expressed genes

Answer 1

[jared.andrews07]

There's probably not going to be a hard and fast answer here. Generally people will implement (very) low-level cutoffs to remove genes that are very lowly or not expressed in all samples, like if there are less than 10 reads for the transcript summed across all samples.

Beyond that, it's difficult to say what's "expressed" versus not. Some genes that are lowly transcribed are still translated into protein at a high level, and vice versa as well. Biology doesn't follow strict rules, and transcripts from different genes are no exception. This is without taking into account variables rates of mRNA degradation/destruction, etc.

Sorry for the non-answer, but you may have to set a (reasonable) cutoff yourself at an arbitrary low-ish value. When people do differential gene expression, this is easier to get around, as the model will generally ignore genes with counts low enough across samples that the statistics don't hold up.