Scalable and Extensible Robinson-Foulds for Comparative Phylogenetics

Robinson-Foulds(RF) is a widely used metric in various phylogenetic analyses including clustering and generating consensus or most-parsimonious trees. Current methods are limited by one or more of the following: 1 versus 1 computation, limited to the basic RF calculation, use one tree collection, are not scalable, and restrict taxa. This paper presents Bipartition Frequency Hash Robinson-Foulds (BFHRF), a scalable and extensible approach for computing the average RF between disparate binary evolutionary tree collections. The novelty of our approach is utilizing a bipartition frequency hash data structure to perform parallelized tree versus hash comparisons in substitution of all possible tree versus tree comparisons. The data structure and updated computation algorithm results in an order of magnitude reduction in both runtime and memory usage. It is 39x faster and 22x reduction in memory compared to HashRF, a fast current method. Additionally, the tree collection distribution can be modified for RF variants and variable taxa due to the lack of restrictions imposed by the hash and retention of all bipartitions. Lastly, BFHRF is implemented in a modular way and provides an easy to use installation and interface for calculating the average RF of query trees against a collection of reference trees. https://github.com/achon/bfhrf