Fast Computations for Measures of Phylogenetic Beta Diversity

For many applications in ecology, it is important to examine the phylogenetic relations between two communities of species. More formally, let T be a phylogenetic tree and let A and B be two samples of its tips, representing the examined communities. We want to compute a value that expresses the phylogenetic diversity between A and B in T. There exist several measures that can do this; these are the so-called phylogenetic beta diversity (β-diversity) measures. Two popular measures of this kind are the Community Distance (CD) and the Common Branch Length (CBL). In most applications, it is not sufficient to compute the value of a beta diversity measure for two communities A and B; we also want to know if this value is relatively large or small compared to all possible pairs of communities in T that have the same size. To decide this, the ideal approach is to compute a standardised index that involves the mean and the standard deviation of this measure among all pairs of species samples that have the same number of elements as A and B. However, no method exists for computing exactly and efficiently this index for CD and CBL. We present analytical expressions for computing the expectation and the standard deviation of CD and CBL. Based on these expressions, we describe efficient algorithms for computing the standardised indices of the two measures. Using standard algorithmic analysis, we provide guarantees on the theoretical efficiency of our algorithms. We implemented our algorithms and measured their efficiency in practice. Our implementations compute the standardised indices of CD and CBL in less than twenty seconds for a hundred pairs of samples on trees with 7 ⋅ 104 tips. Our implementations are available through the R package PhyloMeasures.