Your best option is to normalise your data, transform to logged counts (e.g. regularised log), and then derive Z scores from these. You can then pick a threshold / cut-off for expression / no expression. For example, if you converted your data to Z-scores, you could then check the expression of genes in your condition of interest by setting Z-score cut-offs of:

• >3 'uniquely' expressed
• <-3 'not' expressed

You will find it difficult from RNA-seq data to find genes that are absolutely not expressed (based on how the data is normalised), unless you literally look at those with 0 counts; although, these may have 0 counts for other reasons. If you want to go by this method, then just look at the normalised counts and order them from high-to-low in your condition of interest.

Transcription is a pervasive process and the majority of the transcriptome is transcribed at a low level in most tissues in the body. It's the transcripts that are highly expressed in a particular tissue that can give it its' 'character', though, like MS4A1 (encodes CD20), CD79A, CD79B and B-cells.

Kevin