September 15: Gloor & Reid’s Compositional Data Analysis

This month’s Human Microbiome Journal Club has been cancelled due to an outbreak of vacationitis, but we will resume next month on Friday, September 15 at 3 p.m. in HSC 3N10A.

At the end of September, we’ll be discussing Gloor and Reid’s 2016 paper, Compositional analysis: a valid approach to analyze microbiome high-throughput sequencing data. Gloor and Reid claim that all next-generation sequencing (NGS) data sets are compositional, and that existing ad-hoc statistical tools invented to analyze RNASeq, ChIP-Seq, microbiome, and other NGS data sets should be replaced with more standard tools that account for the data sets’ compositional nature.

Failing to account for the compositionality of a dataset can lead to spurious apparent differential abundances, badly mismeasured correlations among features, and mis-clustering of samples. Gloor and Reid discuss these effects, and propose as a solution the centre-log ratio transformation. ALDEx2 and ANCOM are two R packages discussed in the paper that implement this transformation, and we’ll discuss what they do similarly, what they do differently, and how they compare to DESeq2.

I’ll take us through the paper, and we’ll discuss

1. The effects of compositionality apparent structure
2. The function and motivation of log-ratio transformations and
3. The assumptions this normalization technique implicitly makes about the data, and whether it actually helps us answer the biological questions we’re interested in.

Come join the discussion Friday, September 15 at 3 p.m. in HSC 3N10A.