Chapter 5: Differential Abundance Testing: ANCOM-BC & MaAsLin2
Chapter 5: Differential Abundance Testing: ANCOM-BC & MaAsLin2
Sequencing instruments have fixed throughput capacities. As a result, microbiome data is strictly compositional: total read counts sum to an arbitrary library size, constraining data to a mathematical simplex. If one bacterial species blooms, the relative abundance of all other species artificially drops, even if their absolute biological biomass remains constant.
Applying standard parametric t-tests or naive RNA-seq tools to compositional microbiome tables produces massive false-positive rates (>30–50%). Compositionally aware methods like ANCOM-BC2 or ALDEx2 explicitly correct for sampling fraction bias.
1. Differential Abundance with ANCOM-BC2 in R
library(ANCOMBC)
# Run ANCOM-BC2 with bias correction
out <- ancombc2(
data = physeq_obj,
tax_level = "Genus",
fix_formula = "Treatment + Age + Sex",
p_adj_method = "holm",
pseudo_sens = TRUE,
prv_cut = 0.10, # Filter genera present in <10% of samples
group = "Treatment"
)
# Extract significantly perturbed genera
res <- out$res
sig_taxa <- res[res$diff_TreatmentTreated == TRUE, c("taxon", "lfc_TreatmentTreated", "q_TreatmentTreated")]
print(head(sig_taxa))2. Multivariable Association Modeling with MaAsLin2
MaAsLin2 (Microbiome Multivariable Association with Linear Models) tests associations between microbial features and complex clinical phenotypes while controlling for confounding clinical covariates (e.g. antibiotic usage, BMI, dietary factors).