Stable Isotope Mixing Models with Covariates in R using cosimmr

Emma Govan and Andrew Parnell

2024-05-30

Source: vignettes/cosimmr.Rmd

Introduction

cosimmr is a package in R designed for running Stable Isotope Mixing Models in R, using Fixed Form Variational Bayes. The aim of this vignette is to allow users to quickly learn how to use the package and giving examples of all the features within the package. If you find any bugs in the software, or have new features which you think would be useful, please add this to the Github issues page

Installation of the cosimmr package

First, start Rstudio and find the window with the command prompt (the symbol >). Type

install.packages("cosimmr")

It may ask you to pick your nearest CRAN mirror (the nearest site which hosts R packages). You will then see some activity on the screen as the cosimmr package and the other packages it uses are downloaded. The final line should then read:

package 'cosimmr' successfully unpacked and MD5 sums checked

You then need to load the package. Type

This will load the cosimmr package and all the associated packages. You’ll need to type the library(cosimmr) command every time you start R.

There is some sample data sets (from Inger et al 2006, Nifong et al 2015, Galloway et al 2015) available within cosimmr. Use the following command to access one.

geese_data_day1 = cosimmr::geese_data_day1

This data can then be loaded into cosimmr using the function cosimmr_load

cosimmr_1 <- with(
  geese_data_day1,
  cosimmr_load(
    formula = mixtures ~ 1,
    source_names = source_names,
    source_means = source_means,
    source_sds = source_sds,
    correction_means = correction_means,
    correction_sds = correction_sds,
    concentration_means = concentration_means
  )
)
## Cannot scale when using mixtures ~1

This is a simple example that doesn’t include any covariates. Note the formula is given in the form tracer_data ~ covariates.

An isospace plot can be generated using this data

plot(cosimmr_1)

The data can then be run through cosimmr_ffvb, the main function of the cosimmr package.

cosimmr_1_out = cosimmr_ffvb(cosimmr_1)

Summary statistics for the run can then be viewed, several options are available, such as “statistics” and “quantiles”

summary(cosimmr_1_out, type = "statistics")
## 
## Summary for Observation 1
##                  mean    sd
## P(Zostera)      0.492 0.175
## P(Grass)        0.073 0.021
## P(U.lactuca)    0.241 0.134
## P(Enteromorpha) 0.193 0.131
## sd_d13C_Pl      1.071 0.792
## sd_d15N_Pl      0.806 0.649

The output of this can be plotted, using the plot function. There are several different options available in the plot function.

plot(cosimmr_1_out, type ="prop_histogram", obs = 1)

For another example, with a continuous covariate, we can look at Alligator data from Nifong et al, 2015. For this example we just use Length as the covariate - but the dataset contains other covariates that can be looked at too. First we load in the data into R - this data is included in cosimmr:

data("alligator_data")

Then we use cosimmr_load to create a “cosimmr_in” object

Length = alligator_data$length
cosimmr_ali <-cosimmr_load(
    formula = as.matrix(alligator_data$mixtures) ~ Length,
    source_names = alligator_data$source_names,
    source_means = as.matrix(alligator_data$source_means),
    source_sds = as.matrix(alligator_data$source_sds),
    correction_means = as.matrix(alligator_data$TEF_means),
    correction_sds = as.matrix(alligator_data$TEF_sds))

We then plot our data to make sure our iso-space plot looks good

plot(cosimmr_ali, colour_by_cov = TRUE, cov_name = "Length")

Then we can run the mixing model:

cosimmr_ali_out = cosimmr_ffvb(cosimmr_ali)

We can then look at a summary of the data. This defaults to observation 1.

summary(cosimmr_ali_out, type = "statistics")
## 
## Summary for Observation 1
##                mean    sd
## P(Marine)     0.612 0.029
## P(Freshwater) 0.388 0.029
## sd_d13C       1.842 0.740
## sd_d15N       0.608 0.176

We can create plots of our data. This code creates a proportion plot and a histogram plot of beta value for individuals 1 and 2

plot(cosimmr_ali_out, type = c("prop_histogram", "beta_histogram"), obs = c(1,2), cov_name = "Length")

We can use the predict function to predict proportions for individuals of lengths 100, 210, and 203 by creating a data frame and then using the predict function

x_pred = data.frame(length = c(100,210,302))
alli_pred = predict(cosimmr_ali_out, x_pred)

“alli_pred” can be treated like a normal cosimmr_out object - we can get summary values for each individual or we can create plots

summary(alli_pred, obs = c(1,2,3), type = "statistics")
## 
## Summary for Observation 1
##                mean    sd
## P(Marine)     0.219 0.023
## P(Freshwater) 0.781 0.023
## 
## Summary for Observation 2
##                mean   sd
## P(Marine)     0.718 0.03
## P(Freshwater) 0.282 0.03
## 
## Summary for Observation 3
##                mean    sd
## P(Marine)     0.941 0.016
## P(Freshwater) 0.059 0.016
plot(alli_pred, type = "prop_density")

We can create a covariates_plot to show the change in consumption of Freshwater as an individual increases in Length

plot(cosimmr_ali_out, type = "covariates_plot", source = "Freshwater", cov_name = "Length")

Alternatively we can look at the change in both sources on one plot

plot(cosimmr_ali_out, type = "covariates_plot", cov_name = "Length", one_plot = TRUE, n_pred = 100)