The Effects of Normalization, Transformation, and Rarefaction on Clustering of OTU Abundance

Abstract

Introduction Introduction Methods Discussion Data and Code Availability References Metagenomic clustering presents a unique opportunity to associate and understand communities. Working with Operational Taxonomic Units (OTUs), however, often requires a strategy for handling OTUs that may be over or under represented in a given sample, which is thought of as \"erroneous\". PCR amplification, for example, is known to sometimes non-linearly over-represent more common species Gonzalez et al. 2012. Strategies dealing with this bias include Normalization, Rarefaction, and Log Transformation. Here, we examine how methods to handle potential outlier observations affect de novo estimation of groups using both clustering and matrix factorization methods.\n\nWhile log transformations may affect parametric tests ...