{
  "id": "molecular-phylogenetics",
  "title": "Molecular Phylogenetics",
  "lang": "en",
  "source_lang": "en",
  "summary": "Molecular phylogenetics reconstructs evolutionary relationships from DNA, RNA, and protein sequences using statistical models of molecular change.",
  "category": "science",
  "subcategory": "biology",
  "type": "concept",
  "aliases": [
    "Molecular phylogeny reconstruction",
    "Molecular evolutionary systematics"
  ],
  "generated_at": "2026-11-04",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/molecular-phylogenetics/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/molecular-phylogenetics/",
      "json": "https://aiwiki.page/zh/molecular-phylogenetics.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Molecular phylogenetics is the study of evolutionary relationships using molecular evidence, especially sequences of DNA, RNA, and proteins. It forms a major part of phylogenetics, the reconstruction and interpretation of evolutionary history. By comparing inherited molecular characters across organisms, populations, or individual genes, it estimates patterns of common ancestry, the amount of evolutionary change along lineages, and, with additional information, the timing of evolutionary events.\n\nIts central premise is that molecular sequences retain historical information. Descendants inherit sequences from ancestors, while mutation, recombination, gene duplication, and other processes modify them. Similarities therefore contain evidence of shared ancestry, but they are not a direct record of it: the same change can arise independently, earlier changes can be overwritten, and different parts of a genome can have different histories. Molecular phylogenetics consequently combines biological knowledge with explicit models and statistical inference rather than treating sequence similarity as an evolutionary tree.\n\nThe field has transformed biological classification, research on the origin and diversification of life, and the analysis of pathogens, ancient populations, and gene families. Its results depend on the choice of organisms and molecular markers, the identification of comparable sequence positions, the adequacy of evolutionary models, and the distinction between the history of a gene and that of the organisms carrying it."
    },
    {
      "heading": "Historical foundations and the meaning of a molecular tree",
      "text": "The intellectual foundation of molecular phylogenetics is descent with modification, articulated by Charles Darwin in 1859. Before molecular sequences became available, evolutionary relationships were reconstructed mainly from anatomy, development, and fossils. Molecular comparisons introduced a large set of characters that could be examined across organisms with very different forms. Early comparisons of protein sequences, including cytochrome c and hemoglobin, showed that biochemical differences could be interpreted historically rather than merely functionally.\n\nIn the early 1960s, Emile Zuckerkandl and Linus Pauling developed the idea that sequence differences could provide information about evolutionary time. Emanuel Margoliash’s 1963 comparison of cytochrome c sequences was an influential demonstration of molecular evidence for relationships among organisms. Walter Fitch and Margoliash presented a distance-based tree-building method in 1967. Subsequent advances included Motoo Kimura’s neutral theory of molecular evolution in 1968, Joseph Felsenstein’s practical maximum-likelihood framework for sequence phylogenies in 1981, and the introduction of bootstrap assessment into phylogenetic analysis in 1985. These developments connected molecular comparison with explicit accounts of evolutionary change and statistical uncertainty.\n\nA phylogenetic tree represents a hypothesis about ancestry. Its tips usually denote sampled organisms, sequences, or populations; its internal nodes denote inferred ancestral divergences. The branching arrangement is the topology. A clade consists of an ancestor and all its descendants. Two tips that share an immediate ancestral node are sister groups, but neither is thereby the ancestor of the other. Trees of living organisms do not arrange those organisms on a ladder from primitive to advanced: all living tips have continued evolving since their shared ancestors.\n\nBranch lengths have different meanings in different diagrams. In a phylogram, they commonly measure expected substitutions per sequence position. In a chronogram, they represent elapsed time. A cladogram emphasizes branching order without assigning quantitative meaning to branch lengths. These representations cannot be interpreted interchangeably. A long molecular branch may reflect rapid change, a long interval, or both, rather than greater complexity or a more “evolved” condition.\n\nRooting establishes the direction from ancestors toward descendants. Many molecular methods initially estimate an unrooted tree, which expresses relationships without identifying the oldest split. A root can be inferred using an outgroup, a clock assumption, or models that contain directional information. Outgroup rooting depends on correctly placing the outgroup outside the focal group and can be distorted by a highly divergent outgroup. The root of a sampled tree is also the common ancestor relevant to that sample, not necessarily the origin of the broader biological group."
    },
    {
      "heading": "Molecular evidence and the choice of markers",
      "text": "Molecular phylogenetic evidence ranges from a short segment of one gene to thousands of loci distributed across a genome. Common data include nucleotide sequences, translated amino-acid sequences, gene presence and absence, and genomic rearrangements. Each records a different aspect of evolutionary change. Marker choice therefore depends on the age of the divergences, the organisms under study, and the biological question. A sequence useful for distinguishing closely related populations may be too rapidly evolving for relationships hundreds of millions of years old.\n\nRibosomal RNA has been particularly important because it occurs throughout cellular life and contains both conserved and variable regions. Small-subunit ribosomal RNA comparisons helped establish the distinctiveness of Archaea and reorganize ideas about relationships among major cellular lineages. In bacteria, the 16S ribosomal RNA gene remains widely used for identification and broad phylogenetic placement, although it often lacks sufficient resolution to distinguish closely related strains or some species. A single ribosomal marker is not a substitute for genome-wide evidence when evolutionary histories are complex.\n\nOrganelle genomes offer another major source of evidence. Mitochondrial DNA has been widely used in animal population history and species identification, while chloroplast sequences have been central to plant systematics. Their usefulness reflects features such as copy number, practical accessibility, and, in many groups, relatively simple inheritance. These advantages also impose limitations. An organelle genome generally represents one linked hereditary history rather than an independent sample of the whole organismal genome. Hybridization can transfer an organelle lineage between species, and sex-biased dispersal can make organelle patterns differ from nuclear patterns.\n\nProtein-coding genes can be analyzed as nucleotides, codons, or amino acids. Amino-acid comparisons often retain useful signal across deep divergences because some nucleotide substitutions do not alter the encoded residue. Conversely, translating sequences discards synonymous variation that may be valuable for recent divergences. Noncoding regions, introns, conserved genomic elements, and flanking sequences provide other combinations of conservation and variability. Large studies often use target enrichment or whole-genome sequencing to sample many such regions.\n\nThe quantity of sequence is not the same as the quantity of independent historical evidence. Thousands of positions in one nonrecombining region share a genealogy, while loci separated by recombination may represent distinct genealogies. Missing data also require contextual interpretation: incomplete sequences are not automatically unusable, but biased patterns of missingness can leave particular relationships poorly informed. Taxon sampling matters as much as marker sampling because adding strategically chosen lineages can reveal intermediate changes, clarify character interpretation, and break up misleadingly long branches."
    },
    {
      "heading": "Homology, orthology, and sequence alignment",
      "text": "A molecular phylogenetic comparison begins with a claim about homology: the characters being compared derive from corresponding features in a common ancestor. Similarity alone does not establish homology, because functional constraints can produce similar sequences independently, especially in short motifs. Once a gene family has been identified, another distinction becomes essential. Orthologous genes diverged through a speciation event, whereas paralogous genes diverged through gene duplication. Both are homologous, but their histories answer different questions.\n\nIf an ancient duplication produced two gene copies before several species diverged, a tree mixing those copies may primarily reconstruct the duplication rather than the species relationships. Subsequent gene losses can conceal the duplication, leaving each sampled species with one copy that is not necessarily orthologous to the copies retained elsewhere. Orthology assessment therefore uses gene trees, sequence similarity, genomic neighborhood, copy number, and knowledge of duplication and loss. Reciprocal best sequence matches are useful screening tools but are not a universal proof of orthology.\n\nSequence alignment arranges sequences so that columns represent hypothesized positional homology. Substitutions change character states within columns, while insertions and deletions require gaps. Alignment is relatively straightforward among closely related sequences with few length changes. It becomes difficult when sequences are highly divergent, contain repeated domains, or have undergone extensive insertion and deletion. Progressive alignment methods build an alignment through a sequence of comparisons, while iterative methods revise earlier decisions. Their guide trees assist alignment construction but need not be reliable final phylogenies.\n\nBiological structure can improve alignment. Protein-coding DNA can be aligned through its amino-acid translation and then mapped back to codons, helping preserve the reading frame. Ribosomal RNA alignments can incorporate secondary structure because paired positions are constrained to maintain base pairing. Protein domain boundaries and conserved active-site residues provide additional landmarks. Such information is particularly important when a purely numerical alignment score would favor a biologically implausible arrangement. At the same time, imposing an incorrect structural expectation can introduce its own bias.\n\nUncertain regions are sometimes excluded or analyzed under alternative alignments. Excessive trimming, however, can remove genuine phylogenetic signal and leave an unrepresentative subset of sites. Insertions and deletions can also be informative characters rather than mere obstacles, although many standard substitution analyses treat gaps as missing observations. More elaborate methods model sequence insertion, deletion, and substitution together, sometimes estimating alignment and phylogeny jointly. In all cases, alignment uncertainty is part of phylogenetic uncertainty: strong statistical support calculated from a fixed but incorrect alignment does not validate the underlying homology statements."
    },
    {
      "heading": "Models of molecular evolution",
      "text": "A substitution model describes how molecular character states change along branches. It does not reproduce every biochemical event. Instead, it provides a probabilistic approximation linking ancestral and descendant sequences. Most commonly used nucleotide models treat evolution at a site as a continuous-time Markov process: the probability of a future state depends on the present state and elapsed evolutionary distance, not on the full sequence of earlier states. The instantaneous rates are collected in a rate matrix \\(Q\\), and transition probabilities over a branch of length \\(t\\) are given by\n\n\\[P(t)=e^{Qt}.\\]\n\nThe Jukes–Cantor model, introduced in 1969, assumes equal equilibrium frequencies of the four nucleotides and equal rates among all nucleotide substitutions. Under this model, the corrected distance corresponding to an observed fraction \\(p\\) of differing sites is\n\n\\[d=-\\frac{3}{4}\\ln\\left(1-\\frac{4p}{3}\\right).\\]\n\nThis correction illustrates why observed differences underestimate evolutionary change: a site can change more than once, including back to its initial state. As \\(p\\) approaches \\(0.75\\), the estimate becomes increasingly unstable. The formula is not generally appropriate when its assumptions are strongly violated.\n\nMore flexible models distinguish transitions from transversions, allow unequal nucleotide frequencies, or estimate different rates for different substitution classes. The general time-reversible model allows six exchangeability parameters, together with nucleotide frequencies, subject to scaling and identifiability constraints. Protein models use 20 states and often rely on exchangeability patterns estimated from large collections of alignments. Codon models describe changes among sense codons and can distinguish synonymous from nonsynonymous substitutions. Their parameters connect phylogenetic analysis to questions about functional constraint and natural selection, although parameter interpretation depends on the model and the data.\n\nEvolutionary rates also vary among sites. Functionally critical positions may change slowly, while other positions accept many substitutions. A common approximation assigns site rates according to a gamma distribution, usually represented computationally by several discrete rate categories. Other models allow different amino-acid preferences at different sites or changes in constraints across lineages. Partitioned analyses assign separate parameters to subsets such as genes or codon positions, but the subsets may still share some parameters or a common tree.\n\nModel selection evaluates a trade-off between fit and complexity, commonly using information criteria or predictive approaches. Selecting the best model among a limited set does not establish that any candidate is adequate. Standard models often assume independent sites, stable character frequencies across lineages, and similar substitution processes throughout the tree. Violations can produce systematic errors that persist as the dataset grows. Model adequacy therefore concerns whether a fitted model reproduces relevant features of the observations, not merely whether it outperforms simpler alternatives."
    },
    {
      "heading": "Methods for reconstructing trees",
      "text": "Tree reconstruction searches for relationships that explain the observed molecular data under a chosen criterion. The number of possible trees grows extremely rapidly with the number of sampled taxa. For \\(n\\) labeled taxa, the number of fully resolved unrooted bifurcating trees is\n\n\\[(2n-5)!!,\\]\n\nwhere the double factorial multiplies every other positive integer down to 1. Even moderate datasets therefore require efficient algorithms and heuristic searches rather than exhaustive examination of every topology. Different methods summarize the data differently and make different assumptions about how evolutionary history should be inferred.\n\nDistance methods first estimate pairwise evolutionary distances and then construct a tree approximating those distances. Neighbor joining, introduced by Naruya Saitou and Masatoshi Nei in 1987, is a widely used example. It is computationally efficient and can provide useful exploratory trees or starting points for more intensive searches. Its accuracy depends partly on whether the estimated distances approximate an additive tree metric. Reducing an alignment to pairwise distances also discards some information about the joint distribution of character states across taxa.\n\nMaximum parsimony favors trees requiring the fewest character changes under the chosen character treatment. It does not require an explicit probabilistic substitution model in its standard form. Nevertheless, it has assumptions and can fail systematically under some evolutionary conditions. In particular, unrelated lineages with many independent changes can appear to share derived characters. This phenomenon, known as long-branch attraction, was demonstrated theoretically by Felsenstein in 1978. Model-based methods can reduce this problem when their models adequately describe the process, but they are not immune to analogous artifacts under misspecification.\n\nMaximum likelihood seeks the topology, branch lengths, and model parameters that maximize the probability of the observed data. For an alignment \\(D\\), a tree \\(T\\), and parameters \\(\\theta\\), the objective is \\(L(T,\\theta)=P(D\\mid T,\\theta)\\). Felsenstein’s pruning algorithm efficiently calculates likelihoods by summing over possible ancestral states through recursive operations on the tree. Practical programs combine such calculations with branch-length optimization and topology rearrangements. Multiple searches can be important because heuristic procedures may become trapped in different local optima.\n\nBayesian inference combines the likelihood with prior distributions to estimate a posterior distribution over trees and parameters:\n\n\\[P(T,\\theta\\mid D)\\propto P(D\\mid T,\\theta)P(T,\\theta).\\]\n\nMarkov chain Monte Carlo and related methods approximate this distribution when direct calculation is infeasible. Unlike a single best-tree analysis, Bayesian inference naturally produces a collection of sampled trees, but its reliability depends on appropriate priors, adequate exploration, and convergence. Neither likelihood nor Bayesian analysis automatically resolves weak data, incorrect homology, or a biological history that is poorly represented by one bifurcating tree."
    },
    {
      "heading": "Statistical support and sources of uncertainty",
      "text": "A reconstructed tree is an estimate, and uncertainty can concern topology, branch lengths, ancestral states, model parameters, or dates. These uncertainties are related but distinct. A branch may have a precisely estimated length conditional on one topology while the topology itself remains doubtful. Similarly, an apparently well-supported relationship can depend on an incorrect alignment or substitution model. Statistical support measures are therefore interpretable only in relation to the data, assumptions, and inference procedure that produced them.\n\nThe nonparametric bootstrap assesses sensitivity to the sampled character data. In its conventional phylogenetic form, alignment columns are resampled with replacement to generate replicate datasets of the original length. A tree is inferred for each replicate, and support for a clade is the proportion of replicate trees containing it. A bootstrap value of 90% means that the clade appeared in 90% of those replicate analyses; it is not simply a 90% probability that the clade is true. Interpretation is complicated when sites are correlated or when the resampling unit does not match the relevant biological independence.\n\nBayesian posterior clade probabilities have a different formal meaning: they represent the posterior probability of a clade conditional on the specified model, priors, and data. They are not interchangeable with bootstrap percentages. Posterior probabilities can be high under a misspecified model, and an inadequately mixed sampling procedure can give a misleading impression of certainty. Independent sampling runs, effective sample sizes, and comparisons of sampled tree distributions help diagnose computational problems, although no single diagnostic establishes that every relevant part of tree space has been explored.\n\nUncertainty can also be evaluated by comparing constrained hypotheses. Likelihood-based topology tests assess whether the data discriminate among proposed trees, while parametric simulation can examine expected behavior under a fitted model. These tests have particular null hypotheses and selection assumptions; failure to reject a topology is not proof that it is correct. Summary trees require similar care. A consensus tree can display repeatedly recovered clades and collapse unresolved relationships, but a polytomy may reflect insufficient evidence rather than a genuinely simultaneous divergence.\n\nSystematic error deserves separate treatment because collecting more characters does not necessarily reduce it. Compositional bias, unrecognized paralogy, contamination, inappropriate rooting, and poorly modeled rate variation can all generate strongly supported errors. Sensitivity analyses examine whether relationships persist under alternative taxon selections, alignments, partitions, and models. Gene-wise or site-wise concordance measures provide another perspective by describing how widely a relationship is supported across the dataset. A high bootstrap value for a concatenated alignment need not mean that most genes individually favor the same relationship."
    },
    {
      "heading": "Gene histories, species histories, and evolutionary networks",
      "text": "A gene tree describes the ancestry of sampled gene copies; a species tree describes the branching history of populations or species. These need not coincide. During speciation, ancestral genetic variation can persist through successive divergences, and different descendant species may retain different ancestral lineages. This produces incomplete lineage sorting. It is especially likely when successive speciation events are close together in time relative to ancestral effective population size. Consequently, a correctly estimated gene tree can disagree with the species tree without involving analytical error or gene transfer.\n\nCoalescent theory describes the ancestry of sampled gene copies backward in time. In a simple three-species model, suppose the species tree is \\(((A,B),C)\\), and the internal branch has length \\(t\\) in coalescent units. The probability that a gene tree matches the species tree is\n\n\\[P_{\\mathrm{match}}=1-\\frac{2}{3}e^{-t},\\]\n\nwhile each of the two discordant gene-tree topologies has probability\n\n\\[P_{\\mathrm{discordant}}=\\frac{1}{3}e^{-t}.\\]\n\nThese expressions assume a standard multispecies coalescent model without migration. They illustrate how short ancestral intervals can produce substantial discordance even when all gene trees are inferred perfectly.\n\nConcatenation combines loci into one alignment and usually estimates a shared topology. It can be effective when discordance is limited, but it may be misleading in some conditions with extensive lineage sorting. Multispecies coalescent methods instead model variation among gene genealogies. Some estimate species trees from previously inferred gene trees; others jointly analyze sequences and genealogies. Their assumptions matter: gene-tree estimation error, recombination within loci, limited sampling, and unmodeled gene flow can affect results. Even the most common gene-tree topology need not equal the species-tree topology in certain larger-taxon settings.\n\nDuplication and loss create another source of discordance. Gene-tree reconciliation maps inferred duplications, losses, and sometimes transfers onto a proposed species history. This framework is useful for studying the expansion of gene families and the origin of new molecular functions. Horizontal gene transfer, particularly important in microbial evolution, can move genes between lineages that are not in a direct ancestor–descendant relationship. Hybridization and introgression also transfer genetic material across population or species boundaries, potentially affecting small genomic regions or substantial fractions of a genome.\n\nA branching tree remains useful for many questions, but a phylogenetic network may better represent histories containing reticulation. Network analyses introduce additional parameters and often face difficulties distinguishing alternative causes of discordance. For example, both lineage sorting and introgression can produce conflicting gene trees. Their separation relies on patterns across many loci, branch lengths, population sampling, and explicit statistical models. Genomic history is therefore often described at several levels: a population-divergence framework, a distribution of genealogies, and particular episodes of genetic exchange."
    },
    {
      "heading": "Molecular clocks and evolutionary timescales",
      "text": "A molecular clock connects molecular change with elapsed time. Under a strict clock, a lineage accumulates expected substitutions at a constant rate, so a branch’s expected molecular length is the product of rate and duration. For two lineages diverging from a common ancestor, a simplified relation is\n\n\\[d=2rt,\\]\n\nwhere \\(d\\) is their corrected sequence divergence, \\(r\\) is the substitution rate per lineage per unit time, and \\(t\\) is the time since divergence. This expression assumes equal rates in the two lineages and does not remove the need to account for multiple substitutions.\n\nSubstitution rates are not universal constants. They vary among genes, sites, organisms, and timescales because of differences in mutation processes, generation intervals, functional constraints, selection, and other factors. Relaxed-clock models allow rates to vary among branches, either with specified correlations between ancestral and descendant rates or according to distributions that do not impose such correlations. Local-clock models assign common rates to selected groups of branches. The choice among these models influences both inferred dates and their uncertainty.\n\nMolecular branch lengths alone generally identify the product of rate and time, not rate and time separately. Absolute dates require calibrating information. Fossils can provide evidence that a lineage existed by a particular time, but the oldest known fossil is not automatically the exact date at which that lineage originated. Its use depends on identification, geological age, and placement within the tree. Calibrations distinguish crown groups, which include the last common ancestor of living members and its descendants, from more inclusive stem relationships. Confusing these placements can shift inferred ages substantially.\n\nRadiometric dating and stratigraphic evidence constrain fossil ages independently of molecular sequences. Calibration distributions can incorporate minimum ages, justified upper bounds, and uncertainty in fossil interpretation. Tip-dating approaches may integrate fossil taxa and morphological data directly into a time-scaled analysis. Recent sequences collected at known dates provide another source of calibration when sufficient substitutions have accumulated during the sampling interval. This approach is especially important for rapidly evolving viruses and some ancient genetic datasets, but temporal signal must be demonstrated rather than assumed.\n\nDate estimates depend jointly on sequence information, tree topology, rate models, calibration choices, and demographic or diversification assumptions. Multiple calibrations do not automatically guarantee accuracy, particularly if they share systematic errors. In Bayesian dating, interactions among calibration distributions and tree priors can produce effective age distributions different from their individually specified forms. Reported intervals therefore represent uncertainty conditional on the adopted framework, not every possible source of uncertainty. Comparisons between molecular dates and the fossil record are most informative when these distinct assumptions are made explicit."
    },
    {
      "heading": "Applications across biology",
      "text": "Molecular phylogenetics has reshaped taxonomy by testing whether named groups correspond to clades. It can reveal cryptic lineages, identify misleading similarities caused by convergent evolution, and place organisms whose morphology provides few diagnostic characters. Molecular evidence does not by itself settle every species boundary, because different species concepts emphasize reproductive isolation, ecological differentiation, or lineage history. Strong classification commonly integrates molecular results with anatomy, geography, behavior, and other biological evidence rather than replacing those sources with a single sequence threshold.\n\nDeep evolutionary comparisons have been especially influential. Carl Woese and George Fox’s 1977 ribosomal RNA work distinguished the lineage then called archaebacteria from other prokaryotes. The three-domain system proposed by Woese, Otto Kandler, and Mark Wheelis in 1990 formalized Bacteria, Archaea, and Eucarya as major groups. Later phylogenomic analyses have provided substantial evidence that eukaryotes arose from within archaeal diversity, while the details of early cellular relationships remain sensitive to sampling and modeling. Molecular trees also support the endosymbiotic origin of mitochondria and chloroplasts from bacterial lineages.\n\nIn epidemiology, pathogen sequences help investigate introductions, population structure, and the spread of lineages through time and space. Phylogeography connects genealogical relationships with geographic distributions, while phylodynamics relates genetic variation to processes such as population growth and transmission. These analyses were widely applied to HIV, influenza, Ebola virus, and SARS-CoV-2. Their interpretations depend strongly on sampling, evolutionary rates, and epidemiological context. Closely related sequences do not alone establish direct transmission between two individuals, identify its direction, or exclude an unsampled intermediary.\n\nAncient DNA has made extinct populations directly accessible to molecular comparison. Analyses of Neanderthal and Denisovan genomes revealed both divergence and genetic exchange among human lineages, demonstrating why a strictly branching history can be incomplete. Ancient sequences also inform the histories of domesticated animals, crops, and past ecosystems. Their distinctive challenges include fragmentation, chemical damage, environmental contamination, and contamination from modern organisms. Authentication and damage-aware processing are therefore integral to interpreting the resulting phylogenies.\n\nOther applications connect evolutionary history to function and conservation. Gene-family trees help distinguish inherited functions from lineage-specific innovations. Ancestral sequence reconstruction estimates past molecular states and can guide experimental studies of ancient proteins, although inferred sequences carry uncertainty and may be sensitive to the evolutionary model. In conservation, phylogenetic diversity measures quantify how much evolutionary history is represented by sets of species. DNA barcoding and environmental DNA assist identification and biodiversity surveys, but assigning a sample to a reference sequence is distinct from reconstructing the complete history of the organisms involved."
    },
    {
      "heading": "Phylogenomics, computation, and reproducibility",
      "text": "Phylogenomics extends molecular phylogenetic reasoning to large collections of genes or whole genomes. Its advantages include broader genomic representation and the ability to compare conflicting histories rather than relying on a single marker. Its central difficulty is that genome-scale data magnify both useful signal and systematic error. A dataset containing millions of sites can strongly favor an incorrect relationship if orthology is misidentified, evolutionary processes are inadequately modeled, or different histories are forced into an unsuitable common framework.\n\nComputational workflows typically involve sequence quality assessment, assembly or mapping, gene prediction, homology detection, alignment, model specification, tree inference, and evaluation. Each stage can alter the final result. Contaminant sequences may place a sample near the contaminating organism, while assembly artifacts can create false gene copies. Poor annotations can introduce frameshifts or truncate proteins. Automated pipelines make large analyses feasible, but reproducibility requires recording their filtering rules and intermediate decisions, not merely the command used to infer the final tree.\n\nWidely used programs implement different parts of this process. MAFFT provides multiple-sequence alignment methods; RAxML and IQ-TREE perform likelihood-based phylogenetic inference; MrBayes implements Bayesian tree estimation; and BEAST provides Bayesian frameworks for time-scaled evolutionary analysis. ASTRAL estimates species trees from collections of gene trees under a coalescent-oriented framework. These programs are not interchangeable wrappers around one analysis. Their models, search procedures, support measures, input assumptions, and treatment of partitions can differ, so software choice is part of the scientific specification.\n\nLarge datasets create several distinct computational bottlenecks. Likelihood evaluation scales with the number of sites, taxa, character states, and rate categories, while topology search must explore a vast discrete space. Parallel computation, vectorized numerical operations, site-pattern compression, and efficient tree updates reduce costs. Site-heterogeneous models, joint species-tree analyses, and evolutionary networks can be substantially more demanding than simpler single-tree analyses. Computational convenience therefore often influences the feasible model, making it important to distinguish a practical approximation from an established biological property.\n\nReproducible studies preserve sequence accession information, taxon labels, alignments, excluded regions, partition definitions, software versions, settings, random seeds where relevant, and inferred trees with clearly defined branch annotations. Newick is widely used for tree representation, while richer formats can store metadata and alternative trees. Current methodological research addresses how to incorporate alignment uncertainty, recombination, changing evolutionary constraints, and genomic discordance without making inference computationally intractable. A recurring challenge is to determine which relationships are robust across plausible biological models and which remain dependent on a narrow set of analytical choices."
    }
  ],
  "links": [
    "dna",
    "rna",
    "protein",
    "phylogenetics",
    "mutation",
    "charles-darwin",
    "linus-pauling",
    "gene",
    "genome",
    "ribosomal-rna",
    "archaea",
    "bacteria",
    "mitochondrion",
    "chloroplast",
    "codon",
    "amino-acid",
    "natural-selection",
    "bayesian-inference",
    "bootstrap-method",
    "speciation",
    "horizontal-gene-transfer",
    "introgression",
    "molecular-clock",
    "radiometric-dating",
    "taxonomy",
    "eukaryote",
    "endosymbiotic-theory",
    "epidemiology",
    "ancient-dna",
    "environmental-dna"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}