I have similar doubt. Did you solve it?
I run ancombc with my Phyloseq opbject and extract the data, but we are not certain about the normalization:
out = ancombc(phyloseq = phyloseq_obj, formula = "condition_1", p_adj_method = "BH", zero_cut = 0.90, lib_cut = 1000, group = NULL, struc_zero = FALSE, neg_lb = FALSE,tol = 1e-5, max_iter = 100, conserve = FALSE,alpha = 0.05, global = FALSE) res = out$res df <- as.data.frame(cbind(rownames(res$diff_abn), res$diff_abn, res$beta, res$se, res$p_val, res$q_val))
From this point I go back to the counts’ table (feature table) and extract the list of significant microbiome subset.
I think, due to the quantity of zeros, the relative abundance per sample would be biased. In fact the trend of the comparisons are some times completely different from raw data (counts) or relative abundance (% per sample).
I found the authors of this paper suggested several normalization methods: www.nature.com/articles/s41522-020-00160-w It seems that ancom-bc would not need previous normalization, because it already deals with it. However, I think it would be necessary to apply a normalization of the data in order to show plots of the most significant taxon after ancom-bc.
Thank you in advance!
Read more here: Source link