search

HelenaLC / muscat

Multi-sample multi-group scRNA-seq analysis tools
165 stars 33 forks source link

Mixed model approach with paired sample design #109

Closed hayfre closed 2 years ago

hayfre commented 2 years ago

I have a very similar experimental design to the one mentioned in #12. Is it also possible to include the sample pairing information using the mixed model approach?

label | sample_id | group_id C1 | S1 | control C2 | S2 | control T1 | S1 | treated T2 | S2 | treated

If so, the ∼1+group_id+(1|sample_id) model that is fit for each gene would include the pairing information rather than the unique sample labels. Is this an appropriate model to use or should the pairing information be included in some other way?

plger commented 2 years ago

Hi, It's not appropriate to replace the sample_id with the pair id, but it's possible to use both. Within muscat you won't be able to model the "pair effect" as a random effect, but you can model it as a fixed effect while still treating the sample effect as a random variable. To do this, store the pair id in a variable (e.g. pair_id), and then use mmDS with covs="pair_id". This will amount to: ~1+group_id+pair_id+(1|sample_id) Pierre-Luc

hayfre commented 2 years ago

Hi Pierre-Luc, Thank you for your fast and helpful response! I will try it out.