I would like to ask if someone uses two different aligners to produce count matrices and then run the same alogrithm for DEG analysis, would it make sense to find the intersection of the DEGs in order to conclude to the "real" DEGS?

  • 1
    $\begingroup$ Never heard of that. Main problem in RNA-seq I see is underpowered experiments, therefore unreliable statistics. Just because one aligner may and the other ma not assign certain reads to genes the intersection does not necessarily increase confidence, it also increases the false-negative rate. I would stick to standard procedures, take the aligner (or pseudo/selective aligner, kallisto or salmon) you like most and then proceed with established statistical pacakges. $\endgroup$ – ATpoint Nov 11 at 16:10
  • $\begingroup$ Thank you so much for your feedback!Yes that particular problem I am trying to solve..! $\endgroup$ – marilu Nov 11 at 16:17
  • $\begingroup$ Any particular reason you try to come up with that? Is the gene you are interested in not showing the behaviour your expect so now you try to tweak results? $\endgroup$ – ATpoint Nov 11 at 16:25
  • $\begingroup$ No, I am just trying to understand which algorithm to trust and which approach would give the most reliable results.. I am not trying to tweak results.. $\endgroup$ – marilu Nov 11 at 16:42
  • $\begingroup$ I suggest you read papers that actually benchmark aligners, there is much work already done on this. Currently, the field seems to prefer the pseudo- and selective aligners such as kallisto and (my personal favourit) salmon. I would always read existing benchmarks first before trying to come up with personal solutions. $\endgroup$ – ATpoint Nov 11 at 17:00

This sort of procedure is done in tool papers to benchmark results compared to different aligners/mappers. See the salmon or kallisto papers for examples. Using the intersect of DE results from different mappers will end up yielding somewhat conservative results, not unlike using different differential expression packages and taking the intersect of their outputs. In practice you could just take a more conservative p-value cut-off and save the hassle of using multiple tools, the results would end up being similar.

| improve this answer | |
  • $\begingroup$ Yes, I have read in tutorials that the best approach would be to find the intersection of DEGs produced by different algorithms (deseq2, limmavoom, edger) and that s the reason why I wondered if I should the same with the aligner as well.. Thank you for your answer! $\endgroup$ – marilu Nov 12 at 11:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.