sppairs tutorial

This tutorial gives a basic introduction to the features built in to library(sppairs). Before you go any further, please note that sppairs is under active development, so the tools that are available – and the way that they work – may change slightly over time. However, I will update this page as those changes are made, so the examples given below should be reliable.

We begin by loading the required packages (using devtools), along with some example data:

# prepare libraries
library(devtools)
install_github('mjwestgate/sppairs')
library(sppairs)

# then import a site by species matrix for example purposes
library(cooccur)
data(beetles)

# remove rare species (<10% sites)
dataset <-clean.dataset(beetles, cutoff.min=0.1)

Calculating pairwise associations
The first step we have to take is to choose a method for estimating pairwise associations between species (i.e. columns). Sppairs provides some options for this, e.g.:

or.symmetric(dataset[, 1:2]) # Symmetric odds ratio = 28.89
or.asymmetric(dataset[, 1:2]) # Asymmetric odds ratio = 11.45
mutual.information(dataset[, 1:2]) # MI = 0.235
cor(dataset[, 1], dataset[, 2]) # Pearson's correlation (from base R) = 0.62

Once we have chosen a method, we can apply it to all pairs of species in a dataset. For historical reasons, the default setting is to use asymmetric odds ratios as discussed by Lane et al (2014); but we don’t have to be limited to this.

# default is to calculate the symmetric odds ratio
result <-spaa(dataset)

# to calculate any asymmetric statistic, set asymmetric=TRUE
result <-spaa(dataset, method='or.asymmetric', asymmetric=TRUE)

An unusual case is that given by Lane et al. (2014), where we have repeat visits to each site, and therefore wish to use mixed models to investigate pairwise associations:

# create grouping variable for example purposes ONLY
groups <-as.factor(rep(c(1:10), length.out=dim(beetles)[1]))

# calculate odds ratio using mixed models
result <-spaa(dataset, method='or.glmer', random.effect=groups, asymmetric=TRUE)

# repeat the above, but give more information on the result
result <-spaa(dataset, method='or.glmer', random.effect=groups, asymmetric=TRUE, complex=TRUE)

# alternatively, return some raw coefficients from mixed models, rather than odds ratios
result <-spaa(dataset, method='glmer.coef', random.effect=groups, asymmetric=TRUE)

Note that method="or.glmer" will be much slower than other options, and will most likely give you a lot of error messages (passed from lme4); but these issues shouldn’t cause spaa() to fail.

NOTE: As currently implemented, spaa gives results that are marginally different from the example published by Lane et al. For example, comparing spaa(method="or.glmer") to the +ve values in Table 3 of our paper in Ecology and Evolution using the same data shows a strong positive correlation, but not perfect concordance (lm gives: slope = 1.12 with SE= 0.06; intercept = 0.08 with SE= 0.32; adjusted r-squared = 0.87, n = 40). This is appears to be due to differences in calculation of logistic mixed models between GenStat (which we used for the paper) and R.

Plotting
The code that we use to plot spaa results is for exploratory use only; if you want prettier diagrams, I’d recommend you use igraph (the more professional choice) or circleplot (my personal preference, but with fewer options).

# get some data
result <-spaa(dataset, method='glmer.coef', random.effect=groups, asymmetric=TRUE)[, 1:3]

# built-in plot code (very messy)
plot.spaa(result)

# plot using igraph
graph.data <- graph_from_data_frame(result)
plot(graph.data) # see ?igraph.plotting for more options

# plot using circleplot
result[, 3] <- log10(result[, 3]+1) # reduce effect of extreme values on the plot
circleplot(result) # basic
circleplot(result, plot.control=list(line.breaks=c(-2, -0.5, 0, 0.5, 2))) # prettier version
Advertisements

6 thoughts on “sppairs tutorial

  1. Hi Dr. Westgate

    I would like to calculate the pairwise associations between fungi species, but the SPPAIRS package don’t appear into the R-packages menu. When SPPAIRS will be available?

    Thanks

    1. Hi Victor,
      Sppairs is available on GitHub now, so the code at the top of this page should work to install it, assuming that you have library(dev tools) installed. I don’t have any plans to put it on CRAN just yet, though. Let me know if you have any issues.
      Martin

  2. Hi, thank you for this helpful package (and paper!) I think this is just what I was looking for to compare fish species associations in our study. In the paper- I saw that there were confidence intervals reported for the Odds Ratios. Does your R package include the ability to do 95% confidence intervals?

    1. I’m glad you like it! I don’t think we built confidence intervals in to sppairs- but if you can find some code to calculate it you can wrap it in a function and pass that to spaa like you would with lapply or similar. In the meantime I’ll add it to the request list – thanks for the feedback!

  3. That’s weird – I haven’t looked at sppairs for a while, but I just tried installing it using this code:
    devtools::install_github(“mjwestgate/sppairs”)
    and didn’t have a problem. Some quick googling shows this sort of error can happen when GitHub and devtools are out of sync, or there’s a problem on the GitHub end (e.g. https://stackoverflow.com/questions/48938019/git-pull-push-unable-to-access-https-ssl-routines-seem-to-be-down). Maybe just try again at a later time and see if it’s fixed?

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 )

Google+ photo

You are commenting using your Google+ 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 )

w

Connecting to %s