A Stochastic Local Search algorithm for distance-based phylogeny reconstruction

TL;DR

This paper introduces a novel stochastic local search algorithm for phylogeny reconstruction that effectively handles high mutation rates and horizontal transfer, outperforming existing methods in challenging scenarios.

Contribution

A new unified framework combining the four-points condition and Pauplin's formula, leading to a stochastic local search algorithm that improves phylogeny reconstruction accuracy.

Findings

Outperforms state-of-the-art algorithms on artificial data with high mutation rates.

Shows significant improvement in cases of horizontal transfer.

Effectively reduces the impact of non-additive distances.

Abstract

In many interesting cases the reconstruction of a correct phylogeny is blurred by high mutation rates and/or horizontal transfer events. As a consequence a divergence arises between the true evolutionary distances and the differences between pairs of taxa as inferred from available data, making the phylogenetic reconstruction a challenging problem. Mathematically this divergence translates in a loss of additivity of the actual distances between taxa. In distance-based reconstruction methods, two properties of additive distances were extensively exploited as antagonist criteria to drive phylogeny reconstruction: on the one hand a local property of quartets, i.e., sets of four taxa in a tree, the four-points condition; on the other hand a recently proposed formula that allows to write the tree length as a function of the distances between taxa, the Pauplin's formula. Here we introduce a new reconstruction scheme, that exploits in a unified framework both the four-points condition and the Pauplin's formula. We propose, in particular, a new general class of distance-based Stochastic Local Search algorithms, which reduces in a limit case to the minimization of the Pauplin's length. When tested on artificially generated phylogenies our Stochastic Big-Quartet Swapping algorithmic scheme significantly outperforms state-of-art distance-based algorithms in cases of deviation from additivity due to high rate of back mutations. A significant improvement is also observed with respect to the state-of-art algorithms in case of high rate of horizontal transfer.