Robert Lanfear’s research while affiliated with Australian National University and other places

We present a robust phylogenetic inference method, called the trimmed log-likelihood method, which effectively identifies fast-evolving, saturated, or erroneous sites in both simulated and empirical multiple sequence alignments. This method avoids circularity by dynamically identifying and removing sites without relying on an initial tree, allowing the specific sites removed to change as tree topology and branch lengths are estimated. Our analyses demonstrate that this method outperforms existing approaches, such as the Slow-Fast method, Tree Independent Generation of Evolutionary Rate approach, and Le Quesne Probability statistics, by removing fewer sites while still inferring phylogenies with comparable or greater accuracy. Implemented in IQ-TREE2, the trimmed log-likelihood method is user-friendly with a simple command-line interface. However, challenges remain in addressing heterogeneous evolutionary processes including compositional biases, such as GC bias. Despite these challenges, our approach offers a practical solution for improving phylogenetic inference by automatically and dynamically identifying sites to down-weight during phylogenetic analyses. We recommended that researchers compare trees inferred by varying the proportion of down-weighted sites to monitor changes in tree topology and to identify a set of candidate tree topologies for further consideration.

In phylogenetic studies, both partitioned models and mixture models are used to account for heterogeneity in molecular evolution among the sites of DNA sequence alignments. Partitioned models require the user to specify the grouping of sites into subsets, and then assume that each subset of sites can be modelled by a single common process. Mixture models do not require users to pre-specify subsets of sites, and instead calculate the likelihood of every site under every model, while co-estimating the model weights and parameters. While much research has gone into the optimisation of partitioned models by merging user-specified subsets, there has been less attention paid to the optimisation of mixture models for DNA sequence alignments. In this study, we first ask whether a key assumption of partitioned models – that each user-specified subset can be modelled by a single common process – is supported by the data. Having shown that this is not the case, we then design, implement, test, and apply an algorithm, MixtureFinder, to select the optimum number of classes for a mixture model of Q matrices for the standard models of DNA sequence evolution. We show this algorithm performs well on simulated and empirical datasets and suggest that it may be useful for future empirical studies. MixtureFinder is available in IQ-TREE2, and a tutorial for using MixtureFinder can be found here: http://www.iqtree.org/doc/Complex-Models#mixture-models.

As phylogenomic datasets have grown in size, researchers have developed new ways to measure biological variation and to assess statistical support for specific branches. Larger datasets have more sites and loci, and therefore less sampling variance. While we can more accurately measure the mean signal in these datasets, lower sampling variance is often reflected in uniformly high measures of branch support—such as the bootstrap and posterior probability—limiting their utility. Larger datasets have also revealed substantial biological variation in the topologies found across individual loci, such that the single species tree inferred by most phylogenetic methods represents a limited summary of the data for many purposes. In contrast to measures of statistical support, the degree of underlying topological variation among loci should be approximately constant regardless of the size of the dataset. “Concordance factors” and similar statistics have therefore become increasingly important tools in phylogenetics. In this review, we explain why concordance factors should be thought of as descriptors of topological variation rather than as measures of statistical support, and argue that they provide important information about the predictive power of the species tree not contained in measures of support. We review a growing suite of statistics for measuring concordance, compare them in a common framework that reveals their interrelationships, and demonstrate how to calculate them using an example from birds. We also discuss how measures of topological variation might change in the future as we move beyond estimating a single “tree of life” towards estimating the myriad evolutionary histories underlying genomic variation.

Profile mixture models capture distinct biochemical constraints on the amino acid substitution process at different sites in proteins. These models feature a mixture of time-reversible models with a common matrix of exchangeabilities and distinct sets of equilibrium amino acid frequencies known as profiles. Combining the exchangeability matrix with each profile generates the matrix of instantaneous rates of amino acid exchange for that profile. Currently, empirically estimated exchangeability matrices (e.g., the LG matrix) are widely used for phylogenetic inference under profile mixture models. However, these were estimated using a single profile and are unlikely optimal for profile mixture models. Here, we describe the GTRpmix model that allows maximum likelihood estimation of a common exchangeability matrix under any profile mixture model. We show that exchangeability matrices estimated under profile mixture models differ from the LG matrix, dramatically improving model fit and topological estimation accuracy for empirical test cases. Because the GTRpmix model is computationally expensive, we provide two exchangeability matrices estimated from large concatenated phylogenomic-supermatrices to be used for phylogenetic analyses. One, called Eukaryotic Linked Mixture (ELM), is designed for phylogenetic analysis of proteins encoded by nuclear genomes of eukaryotes, and the other, Eukaryotic and Archaeal Linked mixture (EAL), for reconstructing relationships between eukaryotes and Archaea. These matrices, combined with profile mixture models, fit data better and have improved topology estimation relative to the LG matrix combined with the same mixture models. Starting with version 2.3.1, IQ-TREE2 allows users to estimate linked exchangeabilities (i.e. amino acid exchange rates) under profile mixture models.

Profile mixture models capture distinct biochemical constraints on the amino acid substitution process at different sites in proteins. These models feature a mixture of time-reversible models with a common set of amino acid exchange rates (a matrix of exchangeabilities) and distinct sets of equilibrium amino acid frequencies known as profiles. Combining the exchangeability matrix with each profile generates the matrix of instantaneous rates of amino acid exchange for that profile. Currently, empirically estimated exchangeability matrices (e.g., the LG or WAG matrices) are widely used for phylogenetic inference under profile mixture models. However, such matrices were originally estimated using site homogeneous models with a single set of equilibrium amino acid frequencies; therefore unlikely to be optimal for site heterogeneous profile mixture models. Here we describe the GTRpmix model, implemented in IQ-TREE2, that allows maximum likelihood estimation of a common set of exchangeabilities for all site classes under any profile mixture model. We show that exchangeability matrices estimated in the presence of a site-heterogeneous profile mixture model differ markedly from the widely used LG matrix and dramatically improve model fit and topological estimation accuracy for empirical test cases. Because the GTRpmix model is computationally expensive, we provide two exchangeability matrices estimated from large concatenated phylogenomic supermatrices under the C60 profile mixture model that can be used as fixed matrices for phylogenetic analyses. One of these, called Eukaryotic Linked Mixture (ELM), is designed for phylogenetic analysis of proteins encoded by nuclear genomes of eukaryotes, and the other, Eukaryotic and Archeal Linked mixture (EAL), for reconstructing relationships between eukaryotes and Archaea. These matrices when combined with profile mixture models fit data much better and have improved topology estimation relative to the empirical LG matrix combined with the same underlying mixture models. Version v2.3.1 of IQ-TREE2 implementing these models is available at www.iqtree.org .

In phylogenetic studies, both partitioned models and mixture models are used to account for heterogeneity in molecular evolution among the sites of DNA sequence alignments. Partitioned models require the user to specify the grouping of sites into subsets, and then assume that each subset of sites can be modelled by a single common process. Mixture models do not require users to pre-specify subsets of sites, and instead calculate the likelihood of every site under every model, while co-estimating the model weights. While much research has gone into the optimisation of partitioned models by merging user-specified subsets, there has been less attention paid to the optimisation of mixture models for DNA sequence alignments. In this study, we first ask whether a key assumption of partitioned models – that each user-specified subset can be modelled by a single common process – is supported by the data. Having shown that this is not the case, we then design, implement, test, and apply an algorithm, MixtureFinder, to select the optimum number of classes for a mixture model of Q matrices for the standard models of DNA sequence evolution. We show this algorithm performs well on simulated and empirical datasets and suggest that it may be useful for future empirical studies. MixtureFinder is available in IQ-TREE2, and a tutorial for using MixtureFinder can be found here: http://www.iqtree.org/doc/Complex-Models#mixture-models .

Hundreds or thousands of loci are now routinely used in modern phylogenomic studies. Concatenation approaches to tree inference assume that there is a single topology for the entire dataset, but different loci may have different evolutionary histories due to incomplete lineage sorting, introgression, and/or horizontal gene transfer; even single loci may not be treelike due to recombination. To overcome this shortcoming, we introduce an implementation of a multi-tree mixture model that we call MAST. This model extends a prior implementation by Boussau et al. (2009) by allowing users to estimate the weight of each of a set of pre-specified bifurcating trees in a single alignment. The MAST model allows each tree to have its own weight, topology, branch lengths, substitution model, nucleotide or amino acid frequencies, and model of rate heterogeneity across sites. We implemented the MAST model in a maximum-likelihood framework in the popular phylogenetic software, IQ-TREE. Simulations show that we can accurately recover the true model parameters, including branch lengths and tree weights for a given set of tree topologies, under a wide range of biologically realistic scenarios. We also show that we can use standard statistical inference approaches to reject a single-tree model when data are simulated under multiple trees (and vice versa). We applied the MAST model to multiple primate datasets and found that it can recover the signal of incomplete lineage sorting in the Great Apes, as well as the asymmetry in minor trees caused by introgression among several macaque species. When applied to a dataset of four Platyrrhine species for which standard concatenated maximum likelihood and gene tree approaches disagree, we observe that MAST gives the highest weight (i.e. the largest proportion of sites) to the tree also supported by gene tree approaches. These results suggest that the MAST model is able to analyse a concatenated alignment using maximum likelihood, while avoiding some of the biases that come with assuming there is only a single tree. We discuss how the MAST model can be extended in the future.

Motivation: Neighbour-Joining is one of the most widely used distance-based phylogenetic inference methods. However, current implementations do not scale well for datasets with more than 10,000 sequences. Given the increasing pace of generating new sequence data, particularly in outbreaks of emerging diseases, and the already enormous existing databases of sequence data for which NJ is a useful approach, new implementations of existing methods are warranted. Results: Here we present DecentTree, which provides highly optimised and parallel implementations of Neighbour-Joining and several of its variants. DecentTree is designed as a stand-alone application and a header-only library easily integrated with other phylogenetic software (e.g., it is integral in the popular IQ-TREE software). We show that DecentTree shows similar or improved performance over existing software (BIONJ, Quicktree, FastME, and RapidNJ), especially for handling very large alignments. For example, DecentTree is up to 6-fold faster than the fastest existing Neighbour-Joining software (e.g., RapidNJ) when generating a tree of 64,000 SARS-CoV-2 genomes. Availability: DecentTree is open source and freely available at https://github.com/iqtree/decenttree. All code and data used in this analysis are available on Github (https://github.com/asdcid/Comparison-of-neighbour-joining-software). Supplementary information: Supplementary data are available at Bioinformatics online.

Phylogenetics has been foundational to SARS-CoV-2 research and public health policy, assisting in genomic surveillance, contact tracing, and assessing emergence and spread of new variants. However, phylogenetic analyses of SARS-CoV-2 have often relied on tools designed for de novo phylogenetic inference, in which all data are collected before any analysis is performed and the phylogeny is inferred once from scratch. SARS-CoV-2 datasets do not fit this mold. There are currently over 14 million sequenced SARS-CoV-2 genomes in online databases, with tens of thousands of new genomes added every day. Continuous data collection, combined with the public health relevance of SARS-CoV-2, invites an "online" approach to phylogenetics, in which new samples are added to existing phylogenetic trees every day. The extremely dense sampling of SARS-CoV-2 genomes also invites a comparison between likelihood and parsimony approaches to phylogenetic inference. Maximum likelihood (ML) and pseudo-ML methods may be more accurate when there are multiple changes at a single site on a single branch, but this accuracy comes at a large computational cost, and the dense sampling of SARS-CoV-2 genomes means that these instances will be extremely rare because each internal branch is expected to be extremely short. Therefore, it may be that approaches based on maximum parsimony (MP) are sufficiently accurate for reconstructing phylogenies of SARS-CoV-2, and their simplicity means that they can be applied to much larger datasets. Here, we evaluate the performance of de novo and online phylogenetic approaches, as well as ML, pseudo-ML, and MP frameworks for inferring large and dense SARS-CoV-2 phylogenies. Overall, we find that online phylogenetics produces similar phylogenetic trees to de novo analyses for SARS-CoV-2, and that MP optimization with UShER and matOptimize produces equivalent SARS-CoV-2 phylogenies to some of the most popular ML and pseudo-ML inference tools. MP optimization with UShER and matOptimize is thousands of times faster than presently available implementations of ML and online phylogenetics is faster than de novo inference. Our results therefore suggest that parsimony-based methods like UShER and matOptimize represent an accurate and more practical alternative to established maximum likelihood implementations for large SARS-CoV-2 phylogenies and could be successfully applied to other similar datasets with particularly dense sampling and short branch lengths.

Phylogenetic methods are emerging as a useful tool to understand cancer evolutionary dynamics, including tumor structure, heterogeneity, and progression. Most currently used approaches utilize either bulk whole genome sequencing or single-cell DNA sequencing and are based on calling copy number alterations and single nucleotide variants (SNVs). Single-cell RNA sequencing (scRNA-seq) is commonly applied to explore differential gene expression of cancer cells throughout tumor progression. The method exacerbates the single-cell sequencing problem of low yield per cell with uneven expression levels. This accounts for low and uneven sequencing coverage and makes SNV detection and phylogenetic analysis challenging. In this article, we demonstrate for the first time that scRNA-seq data contain sufficient evolutionary signal and can also be utilized in phylogenetic analyses. We explore and compare results of such analyses based on both expression levels and SNVs called from scRNA-seq data. Both techniques are shown to be useful for reconstructing phylogenetic relationships between cells, reflecting the clonal composition of a tumor. Both standardized expression values and SNVs appear to be equally capable of reconstructing a similar pattern of phylogenetic relationship. This pattern is stable even when phylogenetic uncertainty is taken in account. Our results open up a new direction of somatic phylogenetics based on scRNA-seq data. Further research is required to refine and improve these approaches to capture the full picture of somatic evolutionary dynamics in cancer.