As anyone who has read a recent article on microbial ecology knows, the name of the game is ordination plots. Looking for post-hoc patterns in 16S and metagenomics surveys is pretty much par for the course. Depending on your question and statistical inclinations there are a huge variety of ordination plots to choose from; NMDS, PCA, PCOA… not to mention the distance metrics; UniFrac (weighted or unweighted), Bray-Curtis, Jaccard, etc. One approach is to simply run all these analyses (QIIME does this by default) and then to look for patterns.
But how to choose? If one plot shows a pattern and another shows a different pattern, how is a researcher to decide? Statistics is one possible approach, but we all know what Mark Twain said about statistics.
Better than statistics, is a new approach described in this week’s issue of QIIME THERAPY. In “fMRI analysis of ordination plots”, the authors describe a breakthrough approach for rapid and accurate classification of ordination plots for microbial ecology. Instead of arbitrary statistics, the researchers harbor the power of the unconscious human mind. Any human can look at a series of dots and see patterns but computers struggle with this simple task. Ask a 5-year old what shapes they see in an ordination plot and they’ll have ideas. But, as the authors describe, there are problems when scientists are involved… they tend to overthink the process and focus on irrelevant details like axis labels (meaningless in a PCA plot for example). In this work, the researchers circumvent the conscious brain by placing subjects in an fMRI machine and passing images of ordination plots through their field of vision. Patterns are classified on the basis of the strength of the reaction to the image. The authors show that this method is 93% accurate on mock community data and also unveiled patterns in tests data sets that had been missed by statisticians. In addition, this method takes 10% of the time and requires about 25% of the cost of a statistical analysis.
As a proof of principle, the authors applied this methodology to previously analyzed data from the Human Microbiome Project. While generally supporting previous conclusions, the authors also found that weighted UniFrac analysis suggested a novel correlation between the microbiome of the Hallux (big toe) and the microbiome of the non-dominant hand of participants. They plan further research to understand the causality of this connection.
The preprint for the Phinch software paper is now online (one of my Legacy Eisen Lab projects) Please enjoy the PDF on bioRxiv while we patiently wait for the manuscript to go through the peer review process:
Bik, H.M. and Pitch Interactive (2014) Phinch: An interactive, exploratory data visualization framework for –Omic datasets, bioRxiv, doi: http://dx.doi.org/10.1101/009944
If you’re not familiar with this project – Phinch (http://phinch.org) is an open-source framework for visualizing biological data, funded by a grant from the Alfred P. Sloan foundation. This project has been an interdisciplinary collaboration between researchers (driven by myself at UC Davis) and Pitch Interactive, a data visualization studio in Oakland, CA. If you’re interested in loading up some data in this visualization tool, check out our GitHub wiki for full instructions on preparing files and metadata (if you’re already using the QIIME pipeline, you should be ready to go in ~10 minutes…we tried to make it that easy!)
Full citation: Kittelmann S, Seedorf H, Walters WA, Clemente JC, Knight R, et al. (2013) Simultaneous Amplicon Sequencing to Explore Co-Occurrence Patterns of Bacterial, Archaeal and Eukaryotic Microorganisms in Rumen Microbial Communities. PLoS ONE 8(2): e47879. doi:10.1371/journal.pone.0047879
Basically, the paper describes the development and use of a PCR strategy to simultaneously characterize eukaryotic, bacterial and archaeal microbes from samples.
Primers used are summarized in Table 2
The strategy they employ attempt to correct for differences in amplification differences between the different amplicons which should therefore allow better normalization of relative abundance estimates. See results in Figure 2.
Definitely worth a look.
This is from the “Tree of Life Blog”
of Jonathan Eisen, an evolutionary biologist and Open Access advocate
at the University of California, Davis. For short updates, follow me on Twitter.