Differential abundance analysis methods

To identify differentially abundant genes in EVs from their donor bacterium, P. goldsteinii ASF519 genome, we compared EV and donor bacteria groups using DESeq2 ²⁵ and limma-trend ²⁶ on GeTMM normalized counts. In the DESeq2 pipeline, a negative binomial distribution model was fitted to rounded GeTMM for its suitability in overdispersed data, and a “local” model was employed for dispersion estimation. It modeled gene abundance with a Negative Binomial distribution, accounting for biological variability. DESeq2 internally managed variance stabilization, and Wald’s test was used for hypothesis testing. Limma-trend, initially for microarray data, uses linear models with empirical Bayes moderation for log-fold change estimation, ideal for datasets with limited replicates. In the Limma pipeline, a linear model was fitted to compare EV groups with their simulated donor bacterial counterparts, using log₂(GeTMM) values. Empirical Bayes statistics were employed to stabilize variance estimates across genes. Genes were deemed significantly differentially abundant if they exhibited an adjusted p-value (Benjamini-Hochberg method) below 0.05 and a log₂ fold change (GeTMM) above 1. In cases where DESeq2 and limma-trend consistently identified overlapping genes as significantly over/underrepresented, we defined significant differential abundance as genes identified by both methods.