Failed refutations: further comments on parsimony and likelihood methods and their relationship to Popper's degree of corroboration

Kluge's (2001, Syst. Biol. 50:322-330) continued arguments that phylogenetic methods based on the statistical principle of likelihood are incompatible with the philosophy of science described by Karl Popper are based on false premises related to Kluge's misrepresentations of Popper's philosophy. Contrary to Kluge's conjectures, likelihood methods are not inherently verificationist; they do not treat every instance of a hypothesis as confirmation of that hypothesis. The historical nature of phylogeny does not preclude phylogenetic hypotheses from being evaluated using the probability of evidence. The low absolute probabilities of hypotheses are irrelevant to the correct interpretation of Popper's concept termed degree of corroboration, which is defined entirely in terms of relative probabilities. Popper did not advocate minimizing background knowledge; in any case, the background knowledge of both parsimony and likelihood methods consists of the general assumption of descent with modification and additional assumptions that are deterministic, concerning which tree is considered most highly corroborated. Although parsimony methods do not assume (in the sense of entailing) that homoplasy is rare, they do assume (in the sense of requiring to obtain a correct phylogenetic inference) certain things about patterns of homoplasy. Both parsimony and likelihood methods assume (in the sense of implying by the manner in which they operate) various things about evolutionary processes, although violation of those assumptions does not always cause the methods to yield incorrect phylogenetic inferences. Test severity is increased by sampling additional relevant characters rather than by character reanalysis, although either interpretation is compatible with the use of phylogenetic likelihood methods. Neither parsimony nor likelihood methods assess test severity (critical evidence) when used to identify a most highly corroborated tree(s) based on a single method or model and a single body of data; however, both classes of methods can be used to perform severe tests. The assumption of descent with modification is insufficient background knowledge to justify cladistic parsimony as a method for assessing degree of corroboration. Invoking equivalency between parsimony methods and likelihood models that assume no common mechanism emphasizes the necessity of additional assumptions, at least some of which are probabilistic in nature. Incongruent characters do not qualify as falsifiers of phylogenetic hypotheses except under extremely unrealistic evolutionary models; therefore, justifications of parsimony methods as falsificationist based on the idea that they minimize the ad hoc dismissal of falsifiers are questionable. Probabilistic concepts such as degree of corroboration and likelihood provide a more appropriate framework for understanding how phylogenetics conforms with Popper's philosophy of science. Likelihood ratio tests do not assume what is at issue but instead are methods for testing hypotheses according to an accepted standard of statistical significance and for incorporating considerations about test severity. These tests are fundamentally similar to Popper's degree of corroboration in being based on the relationship between the probability of the evidence e in the presence versus absence of the hypothesis h, i.e., between p(e|hb) and p(e|b), where b is the background knowledge. Both parsimony and likelihood methods are inductive in that their inferences (particular trees) contain more information than (and therefore do not follow necessarily from) the observations upon which they are based; however, both are deductive in that their conclusions (tree lengths and likelihoods) follow necessarily from their premises (particular trees, observed character state distributions, and evolutionary models). For these and other reasons, phylogenetic likelihood methods are highly compatible with Karl Popper's philosophy of science and offer several advantages over parsimony methods in this context.     

Weighted parsimony outperforms other methods of phylogenetic inference under models appropriate for morphology

One of the lasting controversies in phylogenetic inference is the degree to which specific evolutionary models should influence the choice of methods. Model‐based approaches to phylogenetic inference (likelihood, Bayesian) are defended on the premise that without explicit statistical models there is no science, and parsimony is defended on the grounds that it provides the best rationalization of the data, while refraining from assigning specific probabilities to trees or character‐state reconstructions. Authors who favour model‐based approaches often focus on the statistical properties of the methods and models themselves, but this is of only limited use in deciding the best method for phylogenetic inference—such decision also requires considering the conditions of evolution that prevail in nature. Another approach is to compare the performance of parsimony and model‐based methods in simulations, which traditionally have been used to defend the use of models of evolution for DNA sequences. Some recent papers, however, have promoted the use of model‐based approaches to phylogenetic inference for discrete morphological data as well. These papers simulated data under models already known to be unfavourable to parsimony, and modelled morphological evolution as if it evolved just like DNA, with probabilities of change for all characters changing in concert along tree branches. The present paper discusses these issues, showing that under reasonable and less restrictive models of evolution for discrete characters, equally weighted parsimony performs as well or better than model‐based methods, and that parsimony under implied weights clearly outperforms all other methods.     

FPGA acceleration of the phylogenetic likelihood function for Bayesian MCMC inference methods

Background  

Likelihood (ML)-based phylogenetic inference has become a popular method for estimating the evolutionary relationships among species based on genomic sequence data. This method is used in applications such as RAxML, GARLI, MrBayes, PAML, and PAUP. The Phylogenetic Likelihood Function (PLF) is an important kernel computation for this method. The PLF consists of a loop with no conditional behavior or dependencies between iterations. As such it contains a high potential for exploiting parallelism using micro-architectural techniques. In this paper, we describe a technique for mapping the PLF and supporting logic onto a Field Programmable Gate Array (FPGA)-based co-processor. By leveraging the FPGA's on-chip DSP modules and the high-bandwidth local memory attached to the FPGA, the resultant co-processor can accelerate ML-based methods and outperform state-of-the-art multi-core processors.     

Penalized likelihood phylogenetic inference: bridging the parsimony-likelihood gap

The increasing diversity and heterogeneity of molecular data for phylogeny estimation has led to development of complex models and model-based estimators. Here, we propose a penalized likelihood (PL) framework in which the levels of complexity in the underlying model can be smoothly controlled. We demonstrate the PL framework for a four-taxon tree case and investigate its properties. The PL framework yields an estimator in which the majority of currently employed estimators such as the maximum-parsimony estimator, homogeneous likelihood estimator, gamma mixture likelihood estimator, etc., become special cases of a single family of PL estimators. Furthermore, using the appropriate penalty function, the complexity of the underlying models can be partitioned into separately controlled classes allowing flexible control of model complexity.     

Phylogeny of the cycads based on multiple single-copy nuclear genes: congruence of concatenated parsimony,likelihood and species tree inference methods

Background and aims

Despite a recent new classification, a stable phylogeny for the cycads has been elusive, particularly regarding resolution of Bowenia, Stangeria and Dioon. In this study, five single-copy nuclear genes (SCNGs) are applied to the phylogeny of the order Cycadales. The specific aim is to evaluate several gene tree–species tree reconciliation approaches for developing an accurate phylogeny of the order, to contrast them with concatenated parsimony analysis and to resolve the erstwhile problematic phylogenetic position of these three genera.

Methods

DNA sequences of five SCNGs were obtained for 20 cycad species representing all ten genera of Cycadales. These were analysed with parsimony, maximum likelihood (ML) and three Bayesian methods of gene tree–species tree reconciliation, using Cycas as the outgroup. A calibrated date estimation was developed with Bayesian methods, and biogeographic analysis was also conducted.

Key Results

Concatenated parsimony, ML and three species tree inference methods resolve exactly the same tree topology with high support at most nodes. Dioon and Bowenia are the first and second branches of Cycadales after Cycas, respectively, followed by an encephalartoid clade (Macrozamia–Lepidozamia–Encephalartos), which is sister to a zamioid clade, of which Ceratozamia is the first branch, and in which Stangeria is sister to Microcycas and Zamia.

Conclusions

A single, well-supported phylogenetic hypothesis of the generic relationships of the Cycadales is presented. However, massive extinction events inferred from the fossil record that eliminated broader ancestral distributions within Zamiaceae compromise accurate optimization of ancestral biogeographical areas for that hypothesis. While major lineages of Cycadales are ancient, crown ages of all modern genera are no older than 12 million years, supporting a recent hypothesis of mostly Miocene radiations. This phylogeny can contribute to an accurate infrafamilial classification of Zamiaceae.     

Effects of nucleotide composition bias on the success of the parsimony criterion in phylogenetic inference

Convergence in nucleotide composition (CNC) in unrelated lineages is a factor potentially affecting the performance of most phylogeny reconstruction methods. Such convergence has deleterious effects because unrelated lineages show similarities due to similar nucleotide compositions and not shared histories. While some methods (such as the LogDet/paralinear distance measure) avoid this pitfall, the amount of convergence in nucleotide composition necessary to deceive other phylogenetic methods has never been quantified. We examined analytically the relationship between convergence in nucleotide composition and the consistency of parsimony as a phylogenetic estimator for four taxa. Our results show that rather extreme amounts of convergence are necessary before parsimony begins to prefer the incorrect tree. Ancillary observations are that (for unweighted Fitch parsimony) transition/transversion bias contributes to the impact of CNC and, for a given amount of CNC and fixed branch lengths, data sets exhibiting substantial site-to-site rate heterogeneity present fewer difficulties than data sets in which rates are homogeneous. We conclude by reexamining a data set originally used to illustrate the problems caused by CNC. Using simulations, we show that in this case the convergence in nucleotide composition alone is insufficient to cause any commonly used methods to fail, and accounting for other evolutionary factors (such as site-to-site rate heterogeneity) can give a correct inference without accounting for CNC.     

Falsifications and corroborations: Karl Popper's influence on systematics

Over the last three decades, the philosophy of Karl Raimund Popper has had a strong influence on the field of systematic biology. Unequivocally, no other philosopher's work has had such an influence during this formative period in systematics. Much, but not all, of the early discourse on Popper and systematics dealt with the philosophical basis of systematics as a science. More recently Popper's work has been discussed in the systematics literature in relation to specific methodologies such as parsimony and maximum likelihood. In this paper, we provide the reader with a concise summary of Popper's ideas relevant to systematics, review the systematic literature invoking or declining Popper's importance to the field, and make a recommendation for the future course of philosophical thinking in systematics. We try to make clear various authors' interpretations of Popper's work and how those interpretations have impacted systematic thought. Although the reader may come away from this review with a clearer idea of Popper's relevance or lack thereof, our primary hope is that the reader will be compelled to question him- or herself about the philosophical basis of the systematic work that he or she does, and to delve into the literature herein cited. We begin by presenting a synopsis of Popper's philosophical views to allow those views to be placed in the context of systematics.     

Efficiencies of fast algorithms of phylogenetic inference under the criteria of maximum parsimony, minimum evolution, and maximum likelihood when a large number of sequences are used

In phylogenetic inference by maximum-parsimony (MP), minimum-evolution (ME), and maximum-likelihood (ML) methods, it is customary to conduct extensive heuristic searches of MP, ME, and ML trees, examining a large number of different topologies. However, these extensive searches tend to give incorrect tree topologies. Here we show by extensive computer simulation that when the number of nucleotide sequences (m) is large and the number of nucleotides used (n) is relatively small, the simple MP or ML tree search algorithms such as the stepwise addition (SA) plus nearest neighbor interchange (NNI) search and the SA plus subtree pruning regrafting (SPR) search are as efficient as the extensive search algorithms such as the SA plus tree bisection-reconnection (TBR) search in inferring the true tree. In the case of ME methods, the simple neighbor-joining (NJ) algorithm is as efficient as or more efficient than the extensive NJ+TBR search. We show that when ME methods are used, the simple p distance generally gives better results in phylogenetic inference than more complicated distance measures such as the Hasegawa-Kishino-Yano (HKY) distance, even when nucleotide substitution follows the HKY model. When ML methods are used, the simple Jukes-Cantor (JC) model of phylogenetic inference generally shows a better performance than the HKY model even if the likelihood value for the HKY model is much higher than that for the JC model. This indicates that at least in the present case, selecting of a substitution model by using the likelihood ratio test or the AIC index is not appropriate. When n is small relative to m and the extent of sequence divergence is high, the NJ method with p distance often shows a better performance than ML methods with the JC model. However, when the level of sequence divergence is low, this is not the case.     

Probabilistic methods outperform parsimony in the phylogenetic analysis of data simulated without a probabilistic model

To understand patterns and processes of the diversification of life, we require an accurate understanding of taxon interrelationships. Recent studies have suggested that analyses of morphological character data using the Bayesian and maximum likelihood Mk model provide phylogenies of higher accuracy compared to parsimony methods. This has proved controversial, particularly studies simulating morphology‐data under Markov models that assume shared branch lengths for characters, as it is claimed this leads to bias favouring the Bayesian or maximum likelihood Mk model over parsimony models which do not explicitly make this assumption. We avoid these potential issues by employing a simulation protocol in which character states are randomly assigned to tips, but datasets are constrained to an empirically realistic distribution of homoplasy as measured by the consistency index. Datasets were analysed with equal weights and implied weights parsimony, and the maximum likelihood and Bayesian Mk model. We find that consistent (low homoplasy) datasets render method choice largely irrelevant, as all methods perform well with high consistency (low homoplasy) datasets, but the largest discrepancies in accuracy occur with low consistency datasets (high homoplasy). In such cases, the Bayesian Mk model is significantly more accurate than alternative models and implied weights parsimony never significantly outperforms the Bayesian Mk model. When poorly supported branches are collapsed, the Bayesian Mk model recovers trees with higher resolution compared to other methods. As it is not possible to assess homoplasy independently of a tree estimate, the Bayesian Mk model emerges as the most reliable approach for categorical morphological analyses.     

Phylogenetic analysis and intraspecific variation: performance of parsimony,likelihood, and distance methods

Intraspecific variation is abundant in all types of systematic characters but is rarely addressed in simulation studies of phylogenetic method performance. We compared the accuracy of 15 phylogenetic methods using simulations to (1) determine the most accurate method(s) for analyzing polymorphic data (under simplified conditions) and (2) test if generalizations about the performance of phylogenetic methods based on previous simulations of fixed (nonpolymorphic) characters are robust to a very different evolutionary model that explicitly includes intraspecific variation. Simulated data sets consisted of allele frequencies that evolved by genetic drift. The phylogenetic methods included eight parsimony coding methods, continuous maximum likelihood, and three distance methods (UPGMA, neighbor joining, and Fitch-Margoliash) applied to two genetic distance measures (Nei's and the modified Cavalli-Sforza and Edwards chord distance). Two sets of simulations were performed. The first examined the effects of different branch lengths, sample sizes (individuals sampled per species), numbers of characters, and numbers of alleles per locus in the eight-taxon case. The second examined more extensively the effects of branch length in the four-taxon, two-allele case. Overall, the most accurate methods were likelihood, the additive distance methods (neighbor joining and Fitch-Margoliash), and the frequency parsimony method. Despite the use of a very different evolutionary model in the present article, many of the results are similar to those from simulations of fixed characters. Similarities include the presence of the "Felsenstein zone," where methods often fail, which suggests that long-branch attraction may occur among closely related species through genetic drift. Differences between the results of fixed and polymorphic data simulations include the following: (1) UPGMA is as accurate or more accurate than nonfrequency parsimony methods across nearly all combinations of branch lengths, and (2) likelihood and the additive distance methods are not positively misled under any combination of branch lengths tested (even when the assumptions of the methods are violated and few characters are sampled). We found that sample size is an important determinant of accuracy and affects the relative success of methods (i.e., distance and likelihood methods outperform parsimony at small sample sizes). Attempts to generalize about the behavior of phylogenetic methods should consider the extreme examples offered by fixed-mutation models of DNA sequence data and genetic-drift models of allele frequencies.     

Relative efficiencies of the maximum parsimony and distance-matrix methods in obtaining the correct phylogenetic tree

The relative efficiencies of the maximum parsimony (MP) and distance-matrix methods in obtaining the correct tree (topology) were studied by using computer simulation. The distance-matrix methods examined are the neighbor-joining, distance-Wagner, Tateno et al. modified Farris, Faith, and Li methods. In the computer simulation, six or eight DNA sequences were assumed to evolve following a given model tree, and the evolutionary changes of the sequences were followed. Both constant and varying rates of nucleotide substitution were considered. From the sequences thus obtained, phylogenetic trees were constructed using the six tree-making methods and compared with the model (true) tree. This process was repeated 300 times for each different set of parameters. The results obtained indicate that when the number of nucleotide substitutions per site is small and a relatively small number of nucleotides are used, the probability of obtaining the correct topology (P1) is generally lower in the MP method than in the distance-matrix methods. The P1 value for the MP method increases with increasing number of nucleotides but is still generally lower than the value for the NJ or DW method. Essentially the same conclusion was obtained whether or not the rate of nucleotide substitution was constant or whether or not a transition bias in nucleotide substitution existed. The relatively poor performance of the MP method for these cases is due to the fact that information from singular sites is not used in this method. The MP method also showed a relatively low P1 value when the model of varying rate of nucleotide substitution was used and the number of substitutions per site was large. However, the MP method often produced cases in which the correct tree was one of several equally parsimonious trees. When these cases were included in the class of "success," the MP method performed better than the other methods, provided that the number of nucleotide substitutions per site was small.     

Inferring phylogenetic networks by the maximum parsimony criterion: a case study

Horizontal gene transfer (HGT) may result in genes whose evolutionary histories disagree with each other, as well as with the species tree. In this case, reconciling the species and gene trees results in a network of relationships, known as the "phylogenetic network" of the set of species. A phylogenetic network that incorporates HGT consists of an underlying species tree that captures vertical inheritance and a set of edges which model the "horizontal" transfer of genetic material. In a series of papers, Nakhleh and colleagues have recently formulated a maximum parsimony (MP) criterion for phylogenetic networks, provided an array of computationally efficient algorithms and heuristics for computing it, and demonstrated its plausibility on simulated data. In this article, we study the performance and robustness of this criterion on biological data. Our findings indicate that MP is very promising when its application is extended to the domain of phylogenetic network reconstruction and HGT detection. In all cases we investigated, the MP criterion detected the correct number of HGT events required to map the evolutionary history of a gene data set onto the species phylogeny. Furthermore, our results indicate that the criterion is robust with respect to both incomplete taxon sampling and the use of different site substitution matrices. Finally, our results show that the MP criterion is very promising in detecting HGT in chimeric genes, whose evolutionary histories are a mix of vertical and horizontal evolution. Besides the performance analysis of MP, our findings offer new insights into the evolution of 4 biological data sets and new possible explanations of HGT scenarios in their evolutionary history.     

LVB: parsimony and simulated annealing in the search for phylogenetic trees

The program LVB seeks parsimonious phylogenies from nucleotide alignments, using the simulated annealing heuristic. LVB runs fast and gives high quality results. AVAILABILITY: The software is available at http://www.rubic.reading.ac.uk/lvb/ Supplementary information: Supplementary information may be downloaded from http://www.rubic.reading.ac.uk/~daniel/     

Statistical consistency and phylogenetic inference: a brief review

The claim that parsimony can be statistically inconsistent remains the chief criticism of the cladistic approach, and also the main justification for alternative model‐based approaches such as maximum likelihood and Bayesian inference. Despite its refutation in the 1980s, this persistent myth of parsimony's Achilles’ heel is entrenched in the primary literature, and has metastasized into textbooks, as well. Here, I review historical controversies, and offer three short arguments as to why statistical consistency is not only irrelevant to systematics, but to empirical science in general.     

Probabilistic methods surpass parsimony when assessing clade support in phylogenetic analyses of discrete morphological data

Fossil taxa are critical to inferences of historical diversity and the origins of modern biodiversity, but realizing their evolutionary significance is contingent on restoring fossil species to their correct position within the tree of life. For most fossil species, morphology is the only source of data for phylogenetic inference; this has traditionally been analysed using parsimony, the predominance of which is currently challenged by the development of probabilistic models that achieve greater phylogenetic accuracy. Here, based on simulated and empirical datasets, we explore the relative efficacy of competing phylogenetic methods in terms of clade support. We characterize clade support using bootstrapping for parsimony and Maximum Likelihood, and intrinsic Bayesian posterior probabilities, collapsing branches that exhibit less than 50% support. Ignoring node support, Bayesian inference is the most accurate method in estimating the tree used to simulate the data. After assessing clade support, Bayesian and Maximum Likelihood exhibit comparable levels of accuracy, and parsimony remains the least accurate method. However, Maximum Likelihood is less precise than Bayesian phylogeny estimation, and Bayesian inference recaptures more correct nodes with higher support compared to all other methods, including Maximum Likelihood. We assess the effects of these findings on empirical phylogenies. Our results indicate probabilistic methods should be favoured over parsimony.     

Improvement of distance-based phylogenetic methods by a local maximum likelihood approach using triplets

We introduce a new approach to estimate the evolutionary distance between two sequences. This approach uses a tree with three leaves: two of them correspond to the studied sequences, whereas the third is chosen to handle long-distance estimation. The branch lengths of this tree are obtained by likelihood maximization and are then used to deduce the desired distance. This approach, called TripleML, improves the precision of evolutionary distance estimates, and thus the topological accuracy of distance-based methods. TripleML can be used with neighbor-joining-like (NJ-like) methods not only to compute the initial distance matrix but also to estimate new distances encountered during the agglomeration process. Computer simulations indicate that using TripleML significantly improves the topological accuracy of NJ, BioNJ, and Weighbor, while conserving a reasonable computation time. With randomly generated 24-taxon trees and realistic parameter values, combining NJ with TripleML reduces the number of wrongly inferred branches by about 11% (against 2.6% and 5.5% for BioNJ and Weighbor, respectively). Moreover, this combination requires only about 1.5 min to infer a phylogeny of 96 sequences composed of 1,200 nucleotides, as compared with 6.5 h for FastDNAml on the same machine (PC 466 MHz).