Multi Loci Phylogenetic Analysis with Gene Tree Clustering

TL;DR

This paper demonstrates that the normalized cut clustering framework efficiently and accurately clusters gene trees based on their topological similarities in BHV space, outperforming other methods in phylogenetic analysis.

Contribution

It introduces the application of the normalized cut framework to gene tree clustering using BHV space, showing its superior performance over traditional clustering techniques.

Findings

Ncut accurately clusters gene trees in simulated data.

Ncut outperforms hierarchical clustering and is comparable to k-means.

NJp method shows better performance than MLE.

Abstract

Summary: Both theory and empirical evidence indicate that phylogenies (trees) of different genes (loci) do not display precisely matched topologies. This phylogenetic incongruence is attributed to the reticulated evolutionary history of most species due to meiotic sexual recombination in eukaryotes, or horizontal transfers of genetic materials in prokaryotes. Nonetheless, most genes do display topologically related phylogenies; this implies they form cohesive subsets (clusters). In this work, we compare popular clustering methods, and show how the performance of the normalized cut framework is efficient and statistically accurate when obtaining clusters on the set of gene trees based on the geodesic distance between them over the Billera-Holmes-Vogtmann (BHV) tree space. We proceed to present a computational study on the performance of different clustering methods with and without preprocessing under different distance metrics and using a series of dimension reduction techniques. Results: First, we show using simulated data that indeed the Ncut framework accurately clusters the set of gene trees given a species tree under the coalescent process. We then depict the success of our framework by comparing its performance to other clustering techniques, including k-means and hierarchical clustering. The main computational results can be summarized to the stellar performance of the Ncut framework even without dimension reduction, the similar performance portrayed by k-means and Ncut under most dimension reduction schemes, the utter failure of hierarchical clustering to accurately capture clusters, as well as the significantly better performance of the NJp method, as compared to MLE.