Likelihood and Inconsistency

Parsimony can be inconsistent, but not maximum likelihood—likelihood advocates often say. This difference and conclusions drawn from it have provided the main reasons advanced by likelihoodists against the use of parsimony. Recent statistical research, however, shows that maximum likelihood estimation of phylogenetic trees can become inconsistent in all but the simplest cases, so that under realistic conditions the consistency of maximum likelihood cannot be assured. If likelihoodists wish to dispose of parsimony, they will have to find another argument.     

Inconsistency of maximum parsimony revisited

Felsenstein (1978, Syst. Zool. 27:401-410) showed that the method of maximum parsimony can be inconsistent, i.e., lead to an incorrect result with an infinite amount of data. The situation in which this inconsistency occurs is often called the "Felsenstein zone," the phenomenon also known as "long-branch attraction." Felsenstein derived a sufficient inconsistency condition from a model for four taxa with only two different parameters for the probability of change on the five branches connecting the four taxa. In the present paper, his approach is used to derive the inconsistency condition of maximum parsimony from the most general model for four taxa, i.e., with five different parameters for the probabilities of change on the five branches and, for the first time, for characters with k states (k = 2, 3, 4, 5, 6, ...) This is used to determine the factors that can cause the inconsistency of maximum parsimony. It is shown that the probability of change on all five branches and the number of character states play a role in causing inconsistency.     

Likelihood, parsimony, and heterogeneous evolution

Evolutionary rates vary among sites and across the phylogenetic tree (heterotachy). A recent analysis suggested that parsimony can be better than standard likelihood at recovering the true tree given heterotachy. The authors recommended that results from parsimony, which they consider to be nonparametric, be reported alongside likelihood results. They also proposed a mixture model, which was inconsistent but better than either parsimony or standard likelihood under heterotachy. We show that their main conclusion is limited to a special case for the type of model they study. Their mixture model was inconsistent because it was incorrectly implemented. A useful nonparametric model should perform well over a wide range of possible evolutionary models, but parsimony does not have this property. Likelihood-based methods are therefore the best way to deal with heterotachy.     

PROBLEMS WITH "SOFT" POLYTOMIES

Abstract— The "soft" assumption attributes polytomies to lack of data, not simultaneous cladogenesis (the "hard" assumption). Most systematists prefer the first interpretation, but most parsimony programs implicitly use the second. Results can thus be inconsistent with initial assumptions. Under certain circumstances that seem especially typical for large data sets treating higher taxa, it may be valid to eliminate both compatible and incompatible polytomous trees from consideration. Consistent treatment of soft polytomies can reduce the ambiguity of cladistic solutions and improve the resolution, and testability, of phylogenetic hypotheses.     

On the best evolutionary rate for phylogenetic analysis

The effect of the evolutionary rate of a gene on the accuracy of phylogeny reconstruction was examined by computer stimulation. The evolutionary rate is measured by the tree length, that is, the expected total number of nucleotide substitutions per site on the phylogeny. DNA sequence data were simulated using both fixed trees with specified branch lengths and random trees with branch lengths generated from a model of cladogenesis. The parsimony and likelihood methods were used for phylogeny reconstruction, and the proportion of correctly recovered branch partitions by each method was estimated. Phylogenetic methods including parsimony appear quite tolerant of multiple substitutions at the same site. The optimum levels of sequence divergence were even higher than upper limits previously suggested for saturation of substitutions, indicating that the problem of saturation may have been exaggerated. Instead, the lack of information at low levels of divergence should be seriously considered in evaluation of a gene's phylogenetic utility, especially when the gene sequence is short. The performance of parsimony, relative to that of likelihood, does not necessarily decrease with the increase of the evolutionary rate.     

A likelihood approach to character weighting and what it tells us about parsimony and compatibility

The statistical framework of maximum likelihood estimation is used to examine character weighting in inferring phylogenies. A simple probabilistic model of evolution is used, in which each character evolves independently among two states, and different lineages evolve independently. When different characters have different known probabilities of change, all sufficiently small, the proper maximum likelihood method of estimating phylogenies is a weighted parsimony method in which the weights are logarithmically related to the rates of change. When rates of change are taken extremely small, the weights become more equal and unweighted parsimony methods are obtained. When it is known that a few characters have very high rates of change and the rest very low rates, but it is not known which characters are the ones having the high rates, the maximum likelihood criterion supports use of compatibility methods. By varying the fraction of characters believed to have high rates of change one obtains a ‘threshold method’ whose behavior depends on the value of a parameter. By altering this parameter the method changes smoothly from being a parsimony method to being a compatibility method. This provides us with a spectrum of intermediates between these methods. These intermediate methods may be of use in analysing real data.     

A likelihood approach to character weighting and what it tells us about parsimony and compatibility

The statistical framework of maximum likelihood estimation is used to examine character weighting in inferring phylogenies. A simple probabilistic model of evolution is used, in which each character evolves independently among two states, and different lineages evolve independently. When different characters have different known probabilities of change, all sufficiently small, the proper maximum likelihood method of estimating phylogenies is a weighted parsimony method in which the weights are logarithmically related to the rates of change. When rates of change are taken extremely small, the weights become more equal and unweighted parsimony methods are obtained.
When it is known that a few characters have very high rates of change and the rest very low rates, but it is not known which characters are the ones having the high rates, the maximum likelihood criterion supports use of compatibility methods. By varying the fraction of characters believed to have high rates of change one obtains a 'threshold method' whose behavior depends on the value of a parameter. By altering this parameter the method changes smoothly from being a parsimony method to being a compatibility method. This provides us with a spectrum of intermediates between these methods. These intermediate methods may be of use in analysing real data.     

A Bayesian perspective on a non-parsimonious parsimony model

Several stochastic models of character change, when implemented in a maximum likelihood framework, are known to give a correspondence between the maximum parsimony method and the method of maximum likelihood. One such model has an independently estimated branch-length parameter for each site and each branch of the phylogenetic tree. This model--the no-common-mechanism model--has many parameters, and, in fact, the number of parameters increases as fast as the alignment is extended. We take a Bayesian approach to the no-common-mechanism model and place independent gamma prior probability distributions on the branch-length parameters. We are able to analytically integrate over the branch lengths, and this allowed us to implement an efficient Markov chain Monte Carlo method for exploring the space of phylogenetic trees. We were able to reliably estimate the posterior probabilities of clades for phylogenetic trees of up to 500 sequences. However, the Bayesian approach to the problem, at least as implemented here with an independent prior on the length of each branch, does not tame the behavior of the branch-length parameters. The integrated likelihood appears to be a simple rescaling of the parsimony score for a tree, and the marginal posterior probability distribution of the length of a branch is dependent upon how the maximum parsimony method reconstructs the characters at the interior nodes of the tree. The method we describe, however, is of potential importance in the analysis of morphological character data and also for improving the behavior of Markov chain Monte Carlo methods implemented for models in which sites share a common branch-length parameter.     

Testing whether ecological factors promote cladogenesis in a group of tiger beetles (Coleoptera: Cicindelidae)

We investigate the role of ecological differentiation in cladogenesis of a monophyletic group of North American tiger beetles, the subgenus Ellipsoptera (genus: Cicindela), by reconstructing their species-level phylogeny from mitochondrial DNA sequences. Observed reconstructions of ecological characters on the phylogeny are compared to those expected under simple null models of no association with cladogenesis. We find no evidence that ecological disparity is associated with either species coexistence, speciation or long-term persistence and/or radiation of lineages. Ecomorphological traits have evolved in response to differences in habitat occupied by species, but without detectable relationship with cladogenesis.     

Maximum parsimony on subsets of taxa

In this paper we investigate mathematical questions concerning the reliability (reconstruction accuracy) of Fitch's maximum parsimony algorithm for reconstructing the ancestral state given a phylogenetic tree and a character. In particular, we consider the question whether the maximum parsimony method applied to a subset of taxa can reconstruct the ancestral state of the root more accurately than when applied to all taxa, and we give an example showing that this indeed is possible. A surprising feature of our example is that ignoring a taxon closer to the root improves the reliability of the method. On the other hand, in the case of the two-state symmetric substitution model, we answer affirmatively a conjecture of Li, Steel and Zhang which states that under a molecular clock the probability that the state at a single taxon is a correct guess of the ancestral state is a lower bound on the reconstruction accuracy of Fitch's method applied to all taxa.     

A New Method of Inference of Ancestral Nucleotide and Amino Acid Sequences

A statistical method was developed for reconstructing the nucleotide or amino acid sequences of extinct ancestors, given the phylogeny and sequences of the extant species. A model of nucleotide or amino acid substitution was employed to analyze data of the present-day sequences, and maximum likelihood estimates of parameters such as branch lengths were used to compare the posterior probabilities of assignments of character states (nucleotides or amino acids) to interior nodes of the tree; the assignment having the highest probability was the best reconstruction at the site. The lysozyme c sequences of six mammals were analyzed by using the likelihood and parsimony methods. The new likelihood-based method was found to be superior to the parsimony method. The probability that the amino acids for all interior nodes at a site reconstructed by the new method are correct was calculated to be 0.91, 0.86, and 0.73 for all, variable, and parsimony-informative sites, respectively, whereas the corresponding probabilities for the parsimony method were 0.84, 0.76, and 0.51, respectively. The probability that an amino acid in an ancestral sequence is correctly reconstructed by the likelihood analysis ranged from 91.3 to 98.7% for the four ancestral sequences.     

The future of phylogeny reconstruction

A new approach to phylogenetic analysis, parsimony jackknifing, uses simple parsimony calculations combined with resampling of characters to arrive at a tree comprising well-supported groups. This is usually much the same as the consensus of most-parsimonious trees found from extensive multiple-tree calculations, but the new method is thousands of times faster, allowing analysis of much larger data matrices, and also provides information on the strength of support for different groups. Jackknife frequencies provide a more reliable assessment of support than do alternative methods, notably "confidence probability" (CP) and T-PTP testing.     

Relative efficiencies of the maximum-likelihood, neighbor-joining, and maximum-parsimony methods when substitution rate varies with site

The relative efficiencies of the maximum-likelihood (ML), neighbor- joining (NJ), and maximum-parsimony (MP) methods in obtaining the correct topology and in estimating the branch lengths for the case of four DNA sequences were studied by computer simulation, under the assumption either that there is variation in substitution rate among different nucleotide sites or that there is no variation. For the NJ method, several different distance measures (Jukes-Cantor, Kimura two- parameter, and gamma distances) were used, whereas for the ML method three different transition/transversion ratios (R) were used. For the MP method, both the standard unweighted parsimony and the dynamically weighted parsimony methods were used. The results obtained are as follows: (1) When the R value is high, dynamically weighted parsimony is more efficient than unweighted parsimony in obtaining the correct topology. (2) However, both weighted and unweighted parsimony methods are generally less efficient than the NJ and ML methods even in the case where the MP method gives a consistent tree. (3) When all the assumptions of the ML method are satisfied, this method is slightly more efficient than the NJ method. However, when the assumptions are not satisfied, the NJ method with gamma distances is slightly better in obtaining the correct topology than is the ML method. In general, the two methods show more or less the same performance. The NJ method may give a correct topology even when the distance measures used are not unbiased estimators of nucleotide substitutions. (4) Branch length estimates of a tree with the correct topology are affected more easily than topology by violation of the assumptions of the mathematical model used, for both the ML and the NJ methods. Under certain conditions, branch lengths are seriously overestimated or underestimated. The MP method often gives serious underestimates for certain branches. (5) Distance measures that generate the correct topology, with high probability, do not necessarily give good estimates of branch lengths. (6) The likelihood-ratio test and the confidence-limit test, in Felsenstein's DNAML, for examining the statistical of branch length estimates are quite sensitive to violation of the assumptions and are generally too liberal to be used for actual data. Rzhetsky and Nei's branch length test is less sensitive to violation of the assumptions than is Felsenstein's test. (7) When the extent of sequence divergence is < or = 5% and when > or = 1,000 nucleotides are used, all three methods show essentially the same efficiency in obtaining the correct topology and in estimating branch lengths.(ABSTRACT TRUNCATED AT 400 WORDS)      

Bayesian Analysis Using a Simple Likelihood Model Outperforms Parsimony for Estimation of Phylogeny from Discrete Morphological Data

Despite the introduction of likelihood-based methods for estimating phylogenetic trees from phenotypic data, parsimony remains the most widely-used optimality criterion for building trees from discrete morphological data. However, it has been known for decades that there are regions of solution space in which parsimony is a poor estimator of tree topology. Numerous software implementations of likelihood-based models for the estimation of phylogeny from discrete morphological data exist, especially for the Mk model of discrete character evolution. Here we explore the efficacy of Bayesian estimation of phylogeny, using the Mk model, under conditions that are commonly encountered in paleontological studies. Using simulated data, we describe the relative performances of parsimony and the Mk model under a range of realistic conditions that include common scenarios of missing data and rate heterogeneity.     

Trait-dependent diversification and the impact of palaeontological data on evolutionary hypothesis testing in New World ratsnakes (tribe Lampropeltini)

For studies investigating trait evolution, there are at least two important questions. First, have traits under consideration influenced cladogenesis and extinction in the group? Second, how do fossil data alter inferences about trait evolution or diversification‐rate dynamics? However, relatively few studies have assessed these questions. Here, we use recently developed methods to test for trait‐dependent diversification in the New World colubrid snake tribe Lampropeltini. We also integrate data from fossil taxa into phylogenetic estimation of evolutionary parameters using a simple Monte Carlo randomization test. These analyses suggest that ecological conditions in temperate regions are tied to higher rates of cladogenesis, but that body size is not related to diversification in the group. We also find that the inclusion of fossil taxa alters absolute estimates of size and the rate of size evolution, but not the overall pattern of ecomorphological diversification, as well as estimates of evolutionary rates, particularly extinction.     

Parsimony overcomes statistical inconsistency with the addition of more data from the same gene

Many authors have demonstrated that the parsimony method of phylogenetic analysis can fail to estimate phylogeny accurately under certain conditions when data follow a model that stipulates homogeneity of the evolutionary process. These demonstrations further show that no matter how much data are added, parsimony will forever exhibit this statistical inconsistency if the additional data have the same distributional properties as the original data. This final component—that the additional data must follow the same distribution as the original data—is crucial to the demonstration. Recent simulations show, however, that if data evolve heterogeneously, parsimony can perform consistently. Here we show, using natural data, that parsimony can overcome inconsistency if new data from the same gene are added to an analysis already exhibiting a condition indistinguishable from inconsistency. © The Willi Hennig Society 2005.     

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.     

Rapid cladogenesis in marine fishes revisited

Lineages that underwent rapid cladogenesis are attractive systems for the study of mechanisms underlying taxonomic, ecological, morphological, and behavioral diversification. Recently developed statistical methods provide insights into historical patterns of diversity and allow distinguishing bursts of cladogenesis from stochastic background rates in the presence of confounding factors such as extinction and incomplete taxon sampling. Here, we compare the dynamics of speciation in several marine fish lineages some of which were previously proposed to have undergone significant changes of cladogenesis through time. We tested for evidence of episodes of rapid cladogenesis using the constant rate and Monte Carlo constant rate tests that are robust to incomplete taxon sampling. These tests employ the statistic gamma to measure the relative position of internal node in a chronogram. For the first time, we conducted a comparative analysis to address the behavior of the statistic under different chronogram-constructing methods (Langley-Fitch, nonparametric rate smoothing, and penalized likelihood). Although estimates of gamma sometimes differ widely among methods, acceptance or rejection of the constant rate model within a particular clade appears to be robust to the choice of method. Bursts of cladogenesis were detected in 14 of 34 studied datasets. Some of these were previously proposed to represent marine fish "radiations," whereas others are identified anew. Our results indicate that the wider application of tree shape methods that are able to detect significantly elevated rates of speciation is useful to more precisely define clades that underwent episodes of rapid cladogenesis in marine fish clades. Contrasting the patterns of phylogenetic diversification in marine fish lineages may facilitate the identification of common evolutionary trajectories versus idiosyncrasies, and ultimately help towards a better understanding of the factors and processes underlying speciation in the marine realm.     

Probability Steiner trees and maximum parsimony in phylogenetic analysis

The phylogenetic tree (PT) problem has been studied by a number of researchers as an application of the Steiner tree problem, a well-known network optimisation problem. Of all the methods developed for phylogenies the maximum parsimony (MP) method is a simple and commonly used method because it relies on directly observable changes in the input nucleotide or amino acid sequences. In this paper we show that the non-uniqueness of the evolutionary pathways in the MP method leads us to consider a new model of PTs. In this so-called probability representation model, for each site a node in a PT is modelled by a probability distribution of nucleotide or amino acid states, and hence the PT at a given site is a probability Steiner tree, i.e. a Steiner tree in a high-dimensional vector space. In spite of the generality of the probability representation model, in this paper we restrict our study to constructing probability phylogenetic trees (PPT) using the parsimony criterion, as well as discussing and comparing our approach with the classical MP method. We show that for a given input set although the optimal topology as well as the total tree length of the PPT is the same as the PT constructed by the classical MP method, the inferred ancestral states and branch lengths are different and the results given by our method provide a plausible alternative to the classical ones.     

A molecular phylogenetic study of ecological diversification in the Australian lizard genus Ctenophorus.

We present phylogenetic analyses of the lizard genus Ctenophorus using 1,639 aligned positions of mitochondrial DNA sequences containing 799 parsimony-informative characters for samples of 22 species of Ctenophorus and 12 additional Australian agamid genera. Sequences from three protein-coding genes (ND1, ND2, and COI) and eight intervening tRNA genes are examined using both parsimony and maximum-likelihood analyses. Species of Ctenophorus form a monophyletic group with Rankinia adelaidensis, which we suggest placing in Ctenophorus. Ecological differentiation among species of Ctenophorus is most evident in the kinds of habitats used for shelter. Phylogenetic analyses suggest that the ancestral condition is to use burrows for shelter, and that habits of sheltering in rocks and shrubs/hummock grasses represent separately derived conditions. Ctenophorus appears to have undergone extensive cladogenesis approximately 10-12 million years ago, with all three major ecological modes being established at that time.