Hello all –
I understand this has been asked, but would it make sense to run per-OTU linear models that include adjustment covariates (e.g., sex) and use the scaled residuals as input for SPLS(DA)? There are a few papers that have done something similar in the context of Spearman correlations, for instance.
Tanks very much.
hi @padimitriu,
Yes, you could, if there is no confounder between your covariate and the treatment effect. Otherwise some of the treatment effect will be lost before it will be input in sPLS-DA (or you may even generate new unwanted variation in the data). The correlation usually arises if the covariate-treatment design is unbalanced
Note: we are currently working on a method to remove batch effects, but only for one covariate: https://www.biorxiv.org/content/10.1101/2020.10.27.358283v1
Kim-Anh
Excelent – thank you for your reply.
Dear developers,
I looked at the manuscript and sPLSDAbatch seems like a great tool that would help me a lot in adjusting for gender in my data! Thus, I wonder whether there is a tutorial for it? On the github I only find instructions on how to install PLSDAbatch.
Your help is highly appreciated!
Cheers, Stef
I’m not super familiar with sPLSDAbatch but after a quick search I’ve found this page which may help