Aggregating Knockoffs for False Discovery Rate Control With an Application to Gut Microbiome Data

This repository provides the data and implementations of the methods described in the paper Aggregating Knockoffs for False Discovery Rate Control With an Application to Gut Microbiome Data by Fang Xie and Johannes Lederer.

Usage

The file lib/AggregatedKnockoffs.R contains the functions AKO and AKO.m to implement the AKO and modified AKO methods described in the paper.

The file lib/dataGen.R is to generate the synthetic data for the simulations. It contains the linear Guassian data and the logistic regression data.

The file lib/stats.R contains the functions stat.ToDoFDRCon, which is to compute the test statistic W described in the paper. The argument stat.type in the function stat.ToDoFDRCon gives four choices max_lambda_lasso, max_lambda_scad, max_lambda_logistic and estimator_logistic, which stand for four different test statistics.

The file lib/myPlot.R contains the function myPlot to display the relationships between target FDR and actual FDR and between target FDR and power.

Simulations

We provide the files Sim_linear_KO_AKO.R and Sim_logistic_KO_AKO.R for comparing the FDR and power of KO with those of our pipeline modified AKO. Sim_linear_KO_AKO.R is designed for the linear regression case and Sim_logistic_KO_AKO.R is designed for the logistic regression case.

Real Data Analysis

The processed data in RealData/ag-cleaned_L2.txt and RealData/ag-cleaned_L6.txt are download from the website of the American Gut Project (http://americangut.org).

The files lib/DataUS012018.R,lib/ImportData.R and SelectIndLinCols.R are used to import and transform the bacteria phyla and BMI data of United State from RealData/ag-cleaned_L2.txt. The files lib/DataUS012018_genus.R and lib/ImportData_genus.R are used to import and transform the bacteria genera and BMI data of United State from RealData/ag-cleaned_L6.txt.

We provide the files AGPAnalysis.R and `AGPAnalysis_genus.R`` to estimate the bacteria phyla and genera that may have an influence on the obesity, respectively.

Repository Authors

• Fang Xie — Postdoc in Statistics, Ruhr-University Bochum — methodology and R implementation
• Johannes Lederer — Professor in Statistics, Ruhr-University Bochum — methodology and R implementation

Acknowledgements

The file lib/ImportData.R refers to the codes of Lun Li and Johannes Lederer.

Reference

Aggregating Knockoffs for False Discovery Rate Control With an Application to Gut Microbiome Data

Cite as "Xie and Lederer, Aggregating Knockoffs for False Discovery Rate Control With an Application to Gut Microbiome Data, arXiv:1907.03807, 2019"