Abstract:
|
Understanding the role of the human microbiome promises comprehensive achievement of many biomedical objectives that have eluded researchers for decades. Yet, statistical analysis of data from microbiome profiling studies continues to present difficulties for statisticians and data analysts. Challenges common to other types of -omics data include high-dimensionality and limited availability of samples. However, in this presentation, we further discuss some of the most pressing problems (and non-problems) that are special to microbiome data. These concerns include issues (and non-issues) related to sparsity, compositionality, structure, batch effects, etc., that are specialized characteristics. We describe possible solutions as well as limitations of the current literature using a combination of benchmark data sets as well as simulations.
|