This folder contains the code for the simulation studies and the applications in Mao et, al (2018). The examples in the paper are generated under R version 3.4.3 and Rcpp version 0.12.13 (2017-09-24). Note that the working directory are default to be each subfolder when running code inside the subfolder.
The American Gut Project collects fecal, oral, skin, and other body site samples from thousands of participants. Lifestyle and dietary data about each participant is also collected, making it possible to study the associations between the human microbiome and various factors. The data is freely available to public through the link ftp://ftp.microbio.me/AmericanGut. The original American Gut sequences and metadata are available at The European Bioinformatics Institute under the accession ERP012803. Cleaned data including the OTU table, the phylogenetic tree and the covariate information about the participants are available at the data portal ftp://ftp.microbio.me/AmericanGut. Licensing information of the data can be found at https://github.com/biocore/American-Gut/blob/master/LICENSE. In our analysis, we use the July 29, 2016 version of the data.
The folder "./Data_process" contains the raw data and the file to process the raw data. The file "97_otus.tree" contains the phylogenetic tree on the 27774 OTUs in the study. The file "ag_fecal.txt" contains the covariate information on each sample. The file "ag_fecal_from_biom.txt" contains the OTU table on all the OTUs. Running the file "Data_process.R" create a dataset containing the OTU table on the top 100 OTUs and the corresponding phylogenetic tree used in the application section. This file can be simply modified to create dataset containing the top 50 and 75 OTUs used for the simulation studies.
The folder "./Simulation" contains the file used to generate data for the simulation studies. R script "sim_s1.R", "sim_s2.R", "sim_s3.R" and "sim_s4.R" perform one round of each simulation scenario, respectively. Running the first three scripts 3000 times with seed 1-3000 and the last script 750 times with seed 1-750 provides the data used for the figures in the simulation section (Section 3) of the paper. The files "nodetest.R", "ntaxa.R", "phyloscan.R" and "phylostructure.R" are used to implement the PhyloScan test in Tang et, al (2018). We appreciate the authors for sharing their code. Typically, it should take less than 3 hours to finish a single simulation round under a certain simulation scenario (we tested the code on a machine with 16GB memory and 2.5 GHz Intel Core i7 processor).
The folder "./Application" contains the code used to generate results in the Appication section. The file "AG_application.R" applies BGCR to compare groups of samples defined by different dietary habits. Directly running this file will create a series of R datasets (each dataset for each dietary habit) that can be used to create the plots and tables in the application section. Computation time for different dietary habits various due to the different sample size. Typically, the code should be able to finish within 5 hours for each diet (we run the analysis on a machine with 16GB memory and 2.5 GHz Intel Core i7 processor).
Mao J., Chen Y. and Ma L. (2018). Bayesian graphical compositional regression for microbiome data. https://arxiv.org/abs/1712.04723.
Tang, Y., Ma, L. and Nicolae, D.L., (2018). A phylogenetic scan test on a dirichlet-tree multinomial model for microbiome data. The Annals of Applied Statistics, 12(1), pp.1-26.