GPTomics/bio-microbiome-differential-abundance

Version Compatibility

Reference examples tested with: DESeq2 1.42+, ggplot2 3.5+, phyloseq 1.46+, scanpy 1.10+

Before using code patterns, verify installed versions match. If versions differ:

• R: packageVersion('<pkg>') then ?function_name to verify parameters

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Differential Abundance Testing

"Find which taxa differ between my groups" -> Identify differentially abundant taxa between experimental conditions using compositionally-aware methods that account for the relative nature of microbiome data.

• R: ALDEx2::aldex() for CLR-transformed Welch's t-test
• R: ANCOMBC::ancombc2() for bias-corrected log-linear models
• R: Maaslin2::Maaslin2() for multivariable association

The Compositionality Problem

Microbiome data is compositional - abundances are relative, not absolute. Standard tests (t-test, DESeq2) can give false positives.

ALDEx2 (Recommended)

Goal: Identify differentially abundant taxa between groups using a compositionally-aware statistical framework.

Approach: Apply CLR transformation with Monte Carlo sampling on the OTU table, run Welch's t-test per taxon, and filter by FDR-corrected p-value and effect size.

library(ALDEx2)
library(phyloseq)

ps <- readRDS('phyloseq_object.rds')
otu <- as.data.frame(otu_table(ps))
if (!taxa_are_rows(ps)) otu <- t(otu)

# Define groups
groups <- sample_data(ps)$Group

# Run ALDEx2 (CLR transformation + Welch's t-test)
aldex_results <- aldex(otu, groups, mc.samples = 128, test = 'welch',
                       effect = TRUE, include.sample.summary = FALSE)

# Filter significant
sig_aldex <- aldex_results[aldex_results$we.eBH < 0.05 & abs(aldex_results$effect) > 1, ]

# Volcano-like plot
aldex.plot(aldex_results, type = 'MW', test = 'welch')

ANCOM-BC2 (Recommended)

library(ANCOMBC)

# Run ANCOM-BC2 with sensitivity analysis
ancom_result <- ancombc2(data = ps, fix_formula = 'Group',
                         p_adj_method = 'BH', pseudo_sens = TRUE,
                         prv_cut = 0.1, lib_cut = 1000,
                         group = 'Group', struc_zero = TRUE)

# Extract results (includes sensitivity analysis)
res_df <- ancom_result$res

# Primary results
sig_ancom <- res_df[res_df$diff_Group == TRUE, ]

# Check sensitivity (passed_ss = passed sensitivity analysis)
robust_hits <- res_df[res_df$diff_Group == TRUE & res_df$passed_ss_Group == TRUE, ]

MaAsLin2

library(Maaslin2)

# Prepare data
features <- as.data.frame(t(otu_table(ps)))
metadata <- as.data.frame(sample_data(ps))

# Run MaAsLin2
maaslin_results <- Maaslin2(
    input_data = features,
    input_metadata = metadata,
    output = 'maaslin2_output',
    fixed_effects = 'Group',
    normalization = 'CLR',
    transform = 'NONE',
    analysis_method = 'LM'
)

# Results in maaslin2_output/all_results.tsv
sig_maaslin <- maaslin_results$results[maaslin_results$results$qval < 0.05, ]

DESeq2 (with caution)

library(DESeq2)
library(phyloseq)

# Convert to DESeq2 (use geometric mean of poscounts)
ps_deseq <- ps
ps_deseq <- prune_samples(sample_sums(ps_deseq) > 1000, ps_deseq)

dds <- phyloseq_to_deseq2(ps_deseq, ~ Group)
dds <- DESeq(dds, test = 'Wald', fitType = 'parametric', sfType = 'poscounts')

res <- results(dds, alpha = 0.05)
sig_deseq <- res[which(res$padj < 0.05 & abs(res$log2FoldChange) > 1), ]

Visualization

library(ggplot2)

# Volcano plot from ALDEx2
ggplot(aldex_results, aes(x = effect, y = -log10(we.eBH))) +
    geom_point(aes(color = we.eBH < 0.05 & abs(effect) > 1), alpha = 0.6) +
    geom_hline(yintercept = -log10(0.05), linetype = 'dashed') +
    geom_vline(xintercept = c(-1, 1), linetype = 'dashed') +
    scale_color_manual(values = c('grey', 'red')) +
    theme_minimal() +
    labs(x = 'Effect Size', y = '-log10(Adjusted P-value)')

Method Comparison

| Method | Handles | Covariates | Speed | Notes | |--------|---------|------------|-------|-------| | ALDEx2 | Compositionality | Limited | Slow | Best for simple designs | | ANCOM-BC2 | Compositionality, zeros, sensitivity | Yes | Medium | Recommended for complex designs | | MaAsLin2 | Compositionality | Yes | Fast | Good for longitudinal | | DESeq2 | Sparsity (less ideal) | Yes | Fast | Use with caution for microbiome |

Related Skills

• diversity-analysis - Identify overall differences first
• differential-expression/deseq2-basics - Similar concepts
• pathway-analysis/go-enrichment - Enrichment of differential taxa