Differential analysis of bacterial RNA seq data

[madillies]

I analyze transcriptomic data that most of the time come from bacteria or cell lineages. In that case, biological variance between independent cultures is quite low, because all the cultures originate from a single clone. In contrast, the different treatments applied to these cells produce important changes in gene expression. This low within-group variability associated with a high between-condition difference situation results in very large lists of genes that show a significant expression difference between the treatments. But it is doubtful that all these genes be relevant with regard to the biological question associated with the experiment. As DESeq2, edgeR, shrinkSeq, ... have been developed to handle true biological variance, I assume that this biological variance is very much under-estimated in the case of bacteria. Applying more stringent significance thresholds is probably not the appropriate solution to this problem.

Anybody has an idea on how to proceed with bacterial data ? Is there a need for a specific dispersion estimate ?

Thank you very much for your help and advice.

[Michael Love]

That's a good point. I think it's not that the biological variance is underestimated, but that the biological variance of the descendants of the clone is estimated, where you would prefer to make inference about a more general population. It seems the only way at this would be to sample from that more general population.

[madillies]

Thank you very much for your comment, and sorry for this very late reply. In the meantime I have interviewed microbiologists to ask them how to access a more general population. The mutation rate of most bacteria is very low, so including clones from different labs in a given experiment would probably lead to the same situation... It seems to be more a biological issue than a statistical one !