2.3.4. Differential feature analysis

Differential feature analysis was performed using LEfSe (Segata et al., 2011) to identify features (genes, pathways, predicted metabolites, and taxa) that had significantly different abundances based on CDI status. For all datasets, the table was normalized with the counts per million (CPM) method. Only features identified as having significantly differential abundance (Kruskal-Wallis, p ≤ 0.05) with a log(LDA) score of at least 2.0 were considered to be enriched, with the exception of the predicted metabolites dataset in which features only had to differ significantly based on Kruskal-Wallis. Levene’s test was performed using the rstatix package (Kassambara, 2020) in R with CPM-normalized values for differential features identified within the active microbial species dataset, as several of the most differential taxa seemed to be highly variable.

SparseDossa2 (Ma et al., 2021) was used to simulate 1,000 tables with various minimum fold change differences for the active microbial taxa and expressed genes datasets in order to assess the corresponding change required to achieve 80% power for LEfSe to detect significant features. Each simulated table consisted of 39 samples, split into two groups of sizes 19 and 20, and was based on a SparseDossa2 model fitted for the respective data type.

For every simulation, the active microbial taxa simulated tables used 17 of the 346 species (5%) that were present in at least 30% of the samples with a minimum average relative abundance of 0.01% to model differential features and 76 of the 1,515 KOs (5%) meeting those requirements were used for each simulation based on the expressed genes dataset. Simulated samples had an average depth of 10,000,000 for both the active microbial taxa and expressed genes datasets. All simulated tables were CPM-normalized prior to being used as input for LEfSe and subject to the same criteria as the observed data to be considered differential. Differential features associated with the expected group were counted as true positives.