amIcompositional: Simple Tests for Compositional Behaviour of High Throughput Data with Common Transformations

Authors

  • Gregory Brian Gloor The University of Western Ontario

DOI:

https://doi.org/10.17713/ajs.v52i4.1617

Abstract

Compositional approaches are beginning to permeate high throughput biomedical sciences in the areas of microbiome, genomics, transcriptomics and proteomics. Yet non-compositional approaches are still commonly observed. Non-compositional approaches are particularly problematic in network analysis based on correlation, ordination and exploratory data analysis based on distance, and differential abundance analysis based on normalization. Here we describe the aIc R package, a simple tool that answers the fundamental question: does the dataset or normalization exhibit compositional artefacts that will skew interpretations when analyzing high throughput biomedical data? The aIc R package includes options for several of the most widely used normalizations and filtering methods. The R package includes tests for subcompositional dominance and coherence along with perturbation and scale invariance. Exploratory analysis is facilitated by an R Shiny app that makes the process simple for those not wishing to use an R console. This simple approach will allow research groups to acknowledge and account for potential artefacts in data analysis resulting in more robust and reliable inferences.

Downloads

Published

2023-07-19

How to Cite

Gloor, G. B. (2023). amIcompositional: Simple Tests for Compositional Behaviour of High Throughput Data with Common Transformations. Austrian Journal of Statistics, 52(4), 180–197. https://doi.org/10.17713/ajs.v52i4.1617

Issue

Section

Special Issue on Compositional Data Analysis (CoDaWork 2022)