DupLoss-2: Improved Phylogenomic Species Tree Inference under Gene Duplication and Loss

Abstract Accurate species tree reconstruction in the presence of widespread gene duplication and loss is a challenging problem in eukaryote phylogenomics. Many phylogenomics methods have been developed over the years to address this challenge; these range from older methods based on gene tree parsimony to newer quartet-based methods. In this work, we introduce improved software for gene tree parsimony-based species tree reconstruction under gene duplication and loss. The new software, DupLoss-2, uses an improved procedure for computing gene losses and is far more accurate and much easier to use than its previous version released over a decade ago. We thoroughly evaluate DupLoss-2 and eight other existing methods, including ASTRAL-Pro, ASTRAL-Pro 2, DISCO-ASTRAL, DISCO-ASTRID, FastMulRFS, and SpeciesRax, using existing benchmarking data and find that DupLoss-2 outperforms all other methods on most of the datasets. It delivers an average of almost 30% reduction in reconstruction error compared to iGTP-Duploss, the previous version of this software, and a 10% reduction compared to the best performing existing method. DupLoss-2 is written in C++ and is freely available open-source.