rdiversity: diversity measurement in R

Note that the following documentation applies only to the development version of the package, which can be installed using: devtools::install_github("boydorr/rdiversity").

rdiversity is a package for R based around a framework for measuring biodiversity using similarity-sensitive diversity measures. It provides functionality for measuring alpha, beta and gamma diversity of metacommunities (e.g. ecosystems) and their constituent subcommunities, where similarity may be defined as taxonomic, phenotypic, genetic, phylogenetic, functional, and so on. It uses the diversity framework described in the arXiv paper arXiv:1404.6520 (q-bio.QM), “How to partition diversity”.

This package has now reached a stable release and is cross-validated against our Julia package Diversity.jl, which is developed independently. Please raise an issue if you find any problems.

Contents

Getting started:

• Installation
• Generating a metacommunity
• Calculating diversity

Special cases:

• Taxonomic diversity
• Phylogenetic diversity
• Genetic diversity
• User defined distance
• User defined similarity

Miscellaneous:

• Additional tools

Installation

To install rdiversity from CRAN, simply run the following from an R console:

install.packages("rdiversity")

Generating a metacommunity

Before calculating diversity a metacommunity object must be created. This object contains all the information needed to calculate diversity. In the following example, we generate a metacommunity (partition) comprising two species (“cows” and “sheep”), and partitioned across three subcommunities (a, b, and c).

# Load the package into R
library(rdiversity)

# Initialise data
partition <- data.frame(a=c(1,1),b=c(2,0),c=c(3,1))
row.names(partition) <- c("cows", "sheep")

The metacommunity() function takes two arguments, partition and similarity. When species are considered completely distinct, an identity matrix is required, which is generated automatically if the similarity argument is missing, as below:

# Generate metacommunity object
meta <- metacommunity(partition = partition)

Note that a warning is displayed when abundances (rather than relative abundances) are entered into the partition argument. Both are acceptable inputs.

When species share some similarity and a similarity matrix is available, then a similarity object (and the metacommunity object) is generated in the following way:

# Initialise similarity matrix
s <- matrix(c(1, 0.5, 0.5, 1), nrow = 2)
row.names(s) <- c("cows", "sheep")
colnames(s) <- c("cows", "sheep")

# Generate similarity object 
s <- similarity(similarity = s, dat_id = "my_taxonomic")

# Generate metacommunity object
meta <- metacommunity(partition = partition, similarity = s)

Alternatively, if a distance matrix is available, then a distance object is generated in the following way:

# Initialise distance matrix
d <- matrix(c(0, 0.7, 0.7, 0), nrow = 2)
row.names(d) <- c("cows", "sheep")
colnames(d) <- c("cows", "sheep")

# Generate distance object
d <- distance(distance = d, dat_id = "my_taxonomic")

# Convert the distance object to similarity object (by means of a linear or exponential transform)
s <- dist2sim(dist = d, transform = "linear")

# Generate metacommunity object
meta <- metacommunity(partition = partition, similarity = s)

Each metacommunity object contains the following slots:

• @type_abundance : the abundance of types within a metacommunity,
  
• @similarity : the pair-wise similarity of types within a metacommunity,
  
• @ordinariness : the ordinariness of types within a metacommunity,
  
• @subcommunity_weights : the relative weights of subcommunities within a metacommunity, and
• @type_weights : the relative weights of types within a metacommunity.

Calculating diversity

Method 1

This method uses a wrapper function to simplify the pipeline and is recommended if only a few measures are being calculated.

A complete list of these functions is shown below:

• raw_sub_alpha() : estimate of naive-community metacommunity diversity
  
• norm_sub_alpha() : similarity-sensitive diversity of subcommunity j in isolation
  
• raw_sub_rho() : redundancy of subcommunity j
  
• norm_sub_rho() : representativeness of subcommunity j
  
• raw_sub_beta() : distinctiveness of subcommunity j
  
• norm_sub_beta() : estimate of effective number of distinct subcommunities
  
• sub_gamma() : contribution per individual toward metacommunity diversity
  
• raw_meta_alpha() : naive-community metacommunity diversity
  
• norm_meta_alpha() : average similarity-sensitive diversity of subcommunities
  
• raw_meta_rho() : average redundancy of subcommunities
  
• norm_meta_rho() : average representativeness of subcommunities
  
• raw_meta_beta() : average distinctiveness of subcommunities
  
• norm_meta_beta() : effective number of distinct subcommunities
  
• meta_gamma() : metacommunity similarity-sensitive diversity

Each of these functions take two arguments, meta (a metacommunity object) and qs (a vector of q values), and output results as a rdiv object. For example, to calculate normalised subcommunity alpha diversity for q=0, q=1, and q=2:

# Initialise data
partition <- data.frame(a=c(1,1),b=c(2,0),c=c(3,1))
row.names(partition) <- c("cows", "sheep")

# Generate a metacommunity object
meta <- metacommunity(partition)

# Calculate diversity
norm_sub_alpha(meta, 0:2)

However, if multiple measures are required and computational efficiency is an issue, then the following method is recommended (the same results are obtained).

Method 2

This method requires that we first calculate the species-level components, by passing a metacommunity object to the appropriate function; raw_alpha(), norm_alpha(), raw_beta(), norm_beta(), raw_rho(), norm_rho(), or raw_gamma(). Subcommunity- and metacommunity-level diversities are calculated using the functions subdiv() and metadiv(). Since both subcommunity and metacommunity diversity measures are transformations of the same species-level component, this method is computationally more efficient.

# Initialise data
partition <- data.frame(a=c(1,1),b=c(2,0),c=c(3,1))
row.names(partition) <- c("cows", "sheep")

# Generate a metacommunity object
meta <- metacommunity(partition)

# Calculate the species-level component for normalised alpha
component <- norm_alpha(meta)

# Calculate normalised alpha at the subcommunity-level 
subdiv(component, 0:2)

# Likewise, calculate normalised alpha at the metacommunity-level 
metadiv(component, 0:2)

In some instances, it may be useful to calculate all subcommunity (or metacommunity) measures. In which case, a metacommunity object may be passed directly to subdiv() or metadiv():

# Calculate all subcommunity diversity measures
subdiv(meta, 0:2)

# Calculate all metacommunity diversity measures
metadiv(meta, 0:2)

Taxonomic diversity

1. Initialise data:

# Taxonomic lookup table
Species <- c("tenuifolium", "asterolepis", "simplex var.grandiflora", "simplex var.ochnacea")
Genus <- c("Protium", "Quararibea", "Swartzia", "Swartzia")
Family <- c("Burseraceae", "Bombacaceae", "Fabaceae", "Fabaceae")
Subclass <- c("Sapindales", "Malvales", "Fabales", "Fabales")
lookup <- cbind.data.frame(Species, Genus, Family, Subclass)

# Partition matrix
partition <- matrix(rep(1, 8), nrow = 4)
colnames(partition) <- LETTERS[1:2]
rownames(partition) <- lookup$Species

and assign values for each taxonomic level:

values <- c(Species = 0, Genus = 1, Family = 2, Subclass = 3, Other = 4)

2. Generate a distance object from a lookup table using the tax2dist() function:

d <- tax2dist(lookup, values)

By default the tax2dist() argument precompute_dist is TRUE, such that a pairwise distance matrix is calculated automatically and is stored in d@distance. If the taxonomy is too large, precompute_dist can be set to FALSE, which enables pairwise taxonomic similarity to be calculated on the fly, in step 4.

3. Convert the distance object to similarity object (by means of a linear or exponential transform) using the dist2sim() function:

s <- dist2sim(d, "linear")

4. Generate a metacommunity object using the metacommunity() function:

meta <- metacommunity(partition, s)

5. Calculate diversity:

meta_gamma(meta, 0:2)

Phylogenetic diversity

Phylogenetic diversity measures can be broadly split into two categories – those that look at the phylogeny as a whole, such as Faith’s (1992) phylogenetic diversity (Faith’s PD), and those that look at pairwise tip distances, such as mean pairwise distance (MPD; Webb, 2000). The framework of measures presented in this package is able to quantify phylogenetic diversity using both of these methods.

Distance-based phylogenetic diversity

1. Initialise data:

# Example data
tree <- ape::rtree(4)
partition <- matrix(1:12, ncol=3)
partition <- partition/sum(partition)

2. Generate a distance matrix using the phy2dist() function:

d <- phy2dist(tree)

By default the phy2dist() argument precompute_dist is TRUE, such that a pairwise distance matrix is calculated automatically and is stored in d@distance. If the taxonomy is too large, precompute_dist can be set to FALSE, which enables pairwise taxonomic similarity to be calculated on the fly, in step 4.

3. Convert the distance object to similarity object (by means of a linear or exponential transform) using the dist2sim() function:

s <- dist2sim(d, "linear")

4. Generate a metacommunity object using the metacommunity() function

meta <- metacommunity(partition, s)

5. Calculate diversity

meta_gamma(meta, 0:2)

Branch-based phylogenetic diversity

1. Initialise data:

tree <- ape::rtree(4)
partition <- matrix(1:12, ncol=3)
partition <- partition/sum(partition)
colnames(partition) <- letters[1:3]
row.names(partition) <- paste0("sp",1:4)
tree$tip.label <- row.names(partition)

2. Generate a similarity object using the phy2branch() function

s <- phy2branch(tree, partition)

3. Generate a metacommunity object using the metacommunity() function

meta <- metacommunity(partition, s)

4. Calculate diversity

meta_gamma(meta, 0:2)

Note that: a metacommunity that was generated using this approach will contain three additional slots:

• @raw_abundance : the relative abundance of terminal species (where types are then considered to be historical species),
• @raw_structure : the length of evolutionary history of each historical species
• @parameters : parameters associated with historical species

Genetic diversity

1. Initialise data:

data(vcfR_dat)
partition <- matrix(c(1,1,1,0,0,0,0,0,0,1,1,1), ncol = 2)
partition <- partition/sum(partition)

2. Generate a distance matrix using the gen2dist() function, data must be of class vcfR:

d <- gen2dist(vcfR_dat)

3. Convert the distance object to a similarity object (by means of a linear or exponential transform) using the dist2sim() function:

s <- dist2sim(d, transform = 'l')

Note: the dist2sim() function contains an optional argument, max_d, which defines the distance at which pairs of individuals have similarity zero. If not supplied this is set to the maximum distance observed in the distance matrix. If comparing different windows on a genome, for example, it is necessary to ensure max_d is the same for each analysis.

4. Generate a metacommunity object using the metacommunity() function:

meta <- metacommunity(partition, s)

5. Calculate diversity, for example beta diversity measures are used to identify trends such as differentiation:

norm_meta_beta(meta, 0:2)

User defined distance

partition <- matrix(sample(6), nrow = 3)
rownames(partition) <- paste0("sp", 1:3)
partition <- partition / sum(partition)

d <- matrix(c(0,.75,1,.75,0,.3,1,.3,0), nrow = 3)
rownames(d) <- paste0("sp", 1:3)
colnames(d) <- paste0("sp", 1:3)
d <- distance(d, "my_taxonomy")
s <- dist2sim(d, "linear")

meta <- metacommunity(partition, s)

User defined similarity

partition <- matrix(sample(6), nrow = 3)
rownames(partition) <- paste0("sp", 1:3)
partition <- partition / sum(partition)

s <- matrix(c(1,.8,0,.8,1,.1,0,.1,1), nrow = 3)
rownames(s) <- paste0("sp", 1:3)
colnames(s) <- paste0("sp", 1:3)
s <- similarity(s, "my_functional")

meta <- metacommunity(partition, s)

Additional tools

repartition()

tree <- ape::rtree(5)
tree$tip.label <- paste0("sp", 1:5)

partition <- matrix(rep(1,10), nrow = 5)
row.names(partition) <- paste0("sp", 1:5)
partition <- partition / sum(partition)
s <- phy2branch(tree, partition)
meta <- metacommunity(partition, s)

new_partition <- matrix(sample(10), nrow = 5)
row.names(new_partition) <- paste0("sp", 1:5)
new_partition <- new_partition / sum(new_partition)

new_meta <- repartition(meta, new_partition)