Microbiome Gym provides 10 statistical tests for the differential abundance analysis of microbiome count data

We provided a unified interface to use any of the analysis methods. The details can be accessed in github

Extensive analysis can be achieved by utilizing the results curated in MicrobiomeGym. We provide two examples to illustrate use cases.

Example 1: Understanding the performance of an existing MWAS methods.
The Wilcoxon ranksum test is arguable the mostly used statistical test used in MWAS. Let’s examine its overall performance for moderate effect size (effect size = 1) and small sample size (n = 24). Assume the observe datasets have moderate sparsity, we can report the performance of Wilcoxon ranksum test using the simulated datasets from Dirichlet multinomial distribution (left, DM_K=2_n=24_1_Kruskal.pdf). However, what is the dataset have large proportion of zero? The right figure (ZINB_K=2_n=24_1_Kruskal.pdf) shows the perform can degrade to an impressive extent. It may be good to consider alternative statistics tests.

Example 2: A new published method (corncob) is reported with good performance. Will it perform well in a similar scenario to Example 1? A quick way to check is from the ROC plot (DM_K=2_n=24_1_corncob.pdf), which shows improvement over the Wilcoxon ranksum test. Note that we do not require to reproduce the simulation reported in the paper, as MicrobiomeGym already curated simulated datasets to help you evaluate methods. Additionally the codes to reproduce the ROC results (or apply to your own datasets) is available to use too.