I have been spurred on to write something about ordination plots given some horrible plots that I’ve seen in the literature, principally concerning those from the authors of QIIME and EMPeror (I do appreciate your work otherwise!).

The issue

When dealing with multivariate data such as that obtained from microbial ecological studies (e.g. 16S counts, etc), a first approach is to look at the beta-diversity. Here, comparisons are made between each and every sample pair using some distance measure (e.g. bray-Curtis, unifrac, etc) which results in a pairwise distance matrix. Now, all these points can be put into n-dimensional space to perfectly show their relationship with each other, but we live in approximately 3 dimensions (maybe up to 8), so can really only view them best in 3D (on a computer) or 2D (on paper). We determine their positions in a number of dimensions, or axes, using ordination approaches (e.g. nMDS or PCoA), which try to give us the ‘best’ picture of the data axis by axis - the first axis showing the ‘most’ of the variation in the data, the next axis showing the next most variation in the data and so on. What we do next is plot these axes.

The problem I have involves the difference between looking at your data/plots on a computer versus what is to be published. On a computer, we can view a 3D figure quite nicely, and which allows us to see more of the data patterns than if we only used a 2D figure. However, on paper, there are always issues printing 3D figures on a 2D medium, and I am to believe that we should have been told to avoid this (in most cases) early on in our undergraduate studies.

Now, there is absolutely no issue with the tools provided by QIIME for viewing your data on a computer and getting a good ‘feel’ for it - that is the most critical! But what do you do for the publication? For what is going to go on paper?

I more than often see a picture of the the 3D plot (Figure 1), and slightly gag each time. The third axis is often (always?) ambiguous, which is further compounded by the fact that the 3D symbols/sphere, while overlapping, do not aid in much more interpretation of the third axis. It baffles me why such figures are allowed to be published, and I worry if reputation is outweighing scientific integrity (thankfully this is a minor issue!).