2 for 1 deal: a response to Waste Not, Want Not and the benchmarking of ANCOM

This week’s microbiome journal club will look at the recent paper by Rob Knight’s group (Weiss et al., 2017) that simultaneously responded to “Waste Not, Want Not” and confirmed ANCOM as an effective differential abundance analysis technique.  More details about these two contributions below.

Questions to be answered and discussed:

  • When to rarefy 16S data
  • The assumptions of various normalisation techniques (DeSeq2, EdgeR, cumulative sum scaling [metagenomeSeq], relative abundance)
  • Is ANCOM an effective tool?
  • Can we trust these simulations?
  • ANCOM versus other tools that adjust for compositional effects: ALDEx2 (Greg Gloor’s group)
  • Potential conflicts of interest in tool development (more users = more funding? – perhaps true of all research outcomes as more citations tends to get more funding)

Come to HSC 3N10A from 3-4pm on Friday, June 23rd to join in the discussion. I will do my best to make this session entertaining so that everyone can have enough energy to actually come to the Phoenix afterwards.

The paper can be found here.

More details:

A response by Rob Knight’s group to “Waste Not, Want Not” finally gets published, three years* after that paper put the very common practice of rarefying (subsampling to common sequencing depth) in microbiome studies into controversial territory. The impact of “Waste Not, Want Not” was very large and created a divide between two groups that develop competitive microbiome analysis pipelines, those behind QIIME (Rob Knight’s group) and those behind the phyloseq R package (Paul McMurdie and Susan Holmes, authors of “Waste Not, Want Not“). McMurdie and Holmes criticise QIIME as a “one size fits all” pipeline, since “it is often encouraged to rarefy counts as a normalizing transformation prior to any/all analysis” (phyloseq FAQ). Weiss et al, 2017 clears up most confusion about the utility of rarefying, suggesting it is the best method to correct for sequencing depth effects when using presence/absence beta diversity metrics such as unweighted UniFrac and binary Jaccard. Weighted metrics such as Bray-Curtis and weighted UniFrac are not affected as much by differences in sequencing depth and can be used after normalizing by relative abundance. Any effect of sequencing depth can be seen by rarefying many times and doing multiple rarefaction PCoA, which is the best way to “Waste Not” and still fix sequencing depth bias.

*This response was originally submitted in October, 2015

The impact of this paper does not stop there, as Weiss and colleagues also benchmarked ANCOM against many other popular differential abundance analysis techniques (DeSeq2, EdgeR, metagenomeSeq, etc.). ANCOM was shown to be the only technique examined to control false discovery rate in the presence of spurious differences associated with the compositional nature of 16S microbiome data. Can we trust their simulations?


This entry was posted in Talks. Bookmark the permalink.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s