CoChat
Sign Up
Cited by 16,873Genomics and Phylogenetic StudiesRNA and protein synthesis mechanismsPlant Diversity and Evolution

IQ-TREE 2: New Models and Efficient Methods for Phylogenetic Inference in the Genomic Era

Open access
Minh, Bui Quang, Schmidt, Heiko A, Chernomor, Olga, Schrempf, Dominik, Woodhams, Michael D, von Haeseler, Arndt, Lanfear, Robert·2020·Molecular Biology and Evolution
DOI 10.1093/molbev/msaa015PMC PMC7182206

Published version of a preprint

Open this paper in CoChat — free accountOpen PDF

Abstract

IQ-TREE (http://www.iqtree.org, last accessed February 6, 2020) is a user-friendly and widely used software package for phylogenetic inference using maximum likelihood. Since the release of version 1 in 2014, we have continuously expanded IQ-TREE to integrate a plethora of new models of sequence evolution and efficient computational approaches of phylogenetic inference to deal with genomic data. Here, we describe notable features of IQ-TREE version 2 and highlight the key advantages over other software.

AI snapshot

Written from the paper’s full text.

TL;DR

IQ-TREE 2 is a major update to the widely used maximum-likelihood phylogenetics software, adding nonreversible substitution models (enabling rooted tree inference), fast likelihood mapping, new tree-search options, terrace-aware partitioned analysis, and parallelized branch and topology tests. Benchmarks on two large alignments show large speedups over TREE-PUZZLE, PhyML, CONSEL, and RAxML for several analyses.

Key findings

  • IQ-TREE 2 supports >200 time-reversible models plus 99 nonreversible DNA Lie Markov models, UNREST for DNA, and a new general nonreversible amino-acid model (NONREV), enabling rooted tree inference.
  • Likelihood mapping: DNA-data set took 282 min in TREE-PUZZLE vs 1 min (1 core) and 21 s (4 cores) in IQ-TREE 2; AA-data set took 99.5 h in TREE-PUZZLE vs 17 min (1 core) and 4.5 min (4 cores).
  • SH-aLRT branch test: DNA-data set 33.9 h in PhyML vs 1.5 min in IQ-TREE 2 (1,300-fold speedup); AA-data set 173.5 h vs 1.6 min (>2,000-fold speedup).
  • Fast tree search on DNA-data set: default 27 min, fast IQ-TREE 82 s, FastTree2 85 s; on AA-data set: default 5.7 h, fast IQ-TREE 13.9 min, FastTree2 4.3 min (FastTree2 faster but produces substantially worse trees per Zhou et al. 2018).
  • Mixture model parameter optimization for LG4X on AA-data set: 1.9 min (IQ-TREE 2), 3 min (RAxML-NG), 17.6 min (PhyML-mixtures).
  • Memory-saving mode (-mem 0.5) halved RAM with 21% runtime increase on DNA-data set (27 → 32.8 min) and 9% on AA-data set (5.7 → 6.2 h).

Methods

  • Software description and benchmarking study; no new empirical biological dataset was generated.
  • Benchmarks used two published alignments on a 2-GHz CPU server: a DNA alignment of 110 vertebrate species and 25,919 sites (Fong et al. 2012) and an amino-acid alignment of 76 metazoan species and 49,388 sites (Whelan et al. 2017).
  • Comparisons against TREE-PUZZLE, PhyML, RAxML-NG, FastTree2, CONSEL, and ParGenes across tree search, likelihood mapping, branch tests, topology tests, and single-locus inference.

Limitations

  • Not stated in the paper as a formal limitations section; the text notes that ParGenes needed less computational time than IQ-TREE 2 for single-locus inference (DNA: 9.17/5.98 min vs 18.6/9.2 min on 2/4 cores; AA: 1.53/0.81 h vs 2.05/1.05 h), though IQ-TREE 2 had higher parallel efficiency (DNA 101%, AA 97.6% vs ParGenes 76.7% and 94.4%).
  • The paper states FastTree2's speed "was accomplished at the cost of producing substantially worse trees," and that IQ-TREE 2's fast search is a trade-off against the default search.
  • Evident from the text: benchmarks are limited to two datasets on one hardware configuration, and the paper is a software announcement without independent validation of the new nonreversible models' accuracy.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1093/molbev/msaa015

Cites

49 of 49 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 14 cited sources checked · 0 supported · 0 flagged

  • Findingown data

    For the DNA-data set, PhyML version 3.3.20190321 took 33.9 h to perform the SH-aLRT, whereas IQ-TREE 2 needed only 1.5 min (1,300-fold speedup).

    “For the DNA-data set, PhyML version 3.3.20190321 (Guindon et al. 2010) took 33.9 h to perform the SH-aLRT, whereas IQ-TREE 2 needed only 1.5 min (1,300-fold speedup).”
    Quote in the paperSubstantial claim
  • Findingcitation

    IQ-TREE 2 shows a higher parallel efficiency (DNA: 101% and AA: 97.6%) than ParGenes (DNA: 76.7% and AA: 94.4%).

    “IQ-TREE 2 shows a higher parallel efficiency (DNA: 101% and AA: 97.6%) (Grama et al. 2003) than ParGenes (DNA: 76.7% and AA: 94.4%)”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Grama et al. 2003

      Not checked in the free run.

  • Findingown data

    ParGenes needs less computational time than IQ-TREE 2.

    “but ParGenes needs less computational time.”
    Quote in the paperNot a substantial claim
  • Findingown data

    IQ-TREE 2 is substantially faster than PhyML-mixtures.

    “IQ-TREE 2 is also substantially faster than PhyML-mixtures.”
    Quote in the paperNot a substantial claim
  • Premisecitation

    Zhou et al. (2018) independently showed that the tree search algorithm in IQ-TREE exhibits good performance in terms of both computing times and likelihood maximization when compared with other popular ML phylogenetics software such as RAxML and PhyML.

    “Zhou et al. (2018) independently showed that the tree search algorithm in IQ-TREE exhibits good performance in terms of both computing times and likelihood maximization when compared with other popular ML phylogenetics software such as RAxML (Stamatakis 2014) and PhyML (Guindon et al. 2010).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Zhou et al. 2018 source

      Not checked in the free run.

  • May combine two claims. 16
  • Classified as probably not a substantial claim. 2

  • May combine two claims.

    Findingown data

    Analysis of the AA-data set took 99.5 h in TREE-PUZZLE, 17 min in IQ-TREE 2 with one CPU core, and 4.5 min using four cores.

    “Similarly, analysis of the AA-data set took 99.5 h in TREE-PUZZLE, 17 min in IQ-TREE 2 with one CPU core, and 4.5 min using four cores.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the AA-data set, the default tree search took 5.7 h, whereas the fast IQ-TREE search and the FastTree2 search took 13.9 and 4.3 min, respectively.

    “For the AA-data set, the default tree search took 5.7 h, whereas the fast IQ-TREE search and the FastTree2 search took 13.9 and 4.3 min, respectively.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the DNA-data set, IQ-TREE 2 took 27 and 32.8 m (21% increase) under the full- and half-memory mode, respectively, whereas the same analysis needed 5.7 and 6.2 h (9% increase) for the AA-data set.

    “For the DNA-data set, IQ-TREE 2 took 27 and 32.8 m (21% increase) under the full- and half-memory mode, respectively, whereas the same analysis needed 5.7 and 6.2 h (9% increase) for the AA-data set.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the AA-data set, CONSEL took 9.4 min and IQ-TREE 2 took 8.7 min with one CPU core and 2.4 min with four CPU cores.

    “For the AA-data set, CONSEL took 9.4 min and IQ-TREE 2 took 8.7 min with one CPU core and 2.4 min with four CPU cores.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For the DNA-data set, tree search under the general time-reversible model took 2.8 min using 4 CPU cores, whereas the UNREST model took 10.7 min.

    “For the DNA-data set (Fong et al. 2012), tree search under the general time-reversible model took 2.8 min using 4 CPU cores, whereas the UNREST model took 10.7 min.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For the DNA-data set, IQ-TREE 2 inferred 168 locus trees in 18.6 and 9.2 min using 2 and 4 CPU cores, respectively; whereas the same analysis in ParGenes took 9.17 and 5.98 min.

    “For the DNA-data set, IQ-TREE 2 inferred 168 locus trees in 18.6 and 9.2 min using 2 and 4 CPU cores, respectively; whereas the same analysis in ParGenes took 9.17 and 5.98 min.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingnone given

    Optimization of the model parameters for the LG4X model on the AA-data set took 1.9 min in IQ-TREE 2, 3 min in RAxML-NG, and 17.6 min in PhyML-mixtures.

    “For example, optimization of the model parameters for the LG4X model on the AA-data set took 1.9 min in IQ-TREE 2, 3 min in RAxML-NG, and 17.6 min in PhyML-mixtures.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the AA-data set, tree search under the protein general time-reversible model took 2.52 h, whereas the NONREV model took 5.8 h.

    “For the AA-data set (Whelan et al. 2017) tree search under the protein general time-reversible model took 2.52 h, whereas the NONREV model took 5.8 h.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the DNA-data set, the default tree search took 27 min, whereas the fast IQ-TREE search and FastTree2 needed 82 and 85 s, respectively.

    “For our example DNA-data set, the default tree search took 27 min, whereas the fast IQ-TREE search and FastTree2 needed 82 and 85 s, respectively.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Application of the original implementation of quartet likelihood mapping in TREE-PUZZLE to the DNA-data set took 282 min, whereas the improved implementation in IQ-TREE 2 required only 1 min using one CPU core and 21 s using four cores.

    “Application of the original implementation of quartet likelihood mapping in TREE-PUZZLE (Schmidt et al. 2002) (i.e., with 10,000 random quartets and exact ML quartet evaluation) to the DNA-data set took 282 min, whereas the improved implementation in IQ-TREE 2 required only 1 min using one CPU core and 21 s using four cores.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the DNA-data set CONSEL took 5 min to test 100 tree topologies, whereas IQ-TREE 2 took 1.8 min with one CPU core and 1.1 min with four CPU cores.

    “For the DNA-data set CONSEL (Shimodaira and Hasegawa 2001), the original and unparallelized implementation of the approximately unbiased test, took 5 min to test 100 tree topologies, whereas IQ-TREE 2 took 1.8 min with one CPU core and 1.1 min with four CPU cores.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    On the AA-data set, PhyML took 173.5 h to perform the SH-aLRT tests, whereas IQ-TREE 2 needed just 1.6 min (a greater than 2,000-fold speedup).

    “On the AA-data set, PhyML took 173.5 h to perform the SH-aLRT tests, whereas IQ-TREE 2 needed just 1.6 min (a greater than 2,000-fold speedup).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the AA-data set with 127 loci, IQ-TREE 2 took 2.05 and 1.05 h using 2 and 4 cores, respectively; whereas ParGenes needed 1.53 and 0.81 h.

    “For the AA-data set with 127 loci, IQ-TREE 2 took 2.05 and 1.05 h using 2 and 4 cores, respectively; whereas ParGenes needed 1.53 and 0.81 h.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    IQ-TREE is a widely used and open-source software package for phylogenetic inference using maximum likelihood.

    “IQ-TREE is a widely used and open-source software package for phylogenetic inference using the maximum likelihood (ML) criterion.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Afgan et al. 2018 source

      Not checked in the free run.

    • cites Hadfield et al. 2018 source

      Not checked in the free run.

    • cites Emms and Kelly 2015

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    IQ-TREE is an integral component of many popular open-source applications such as Galaxy, Nextstrain, OrthoFinder, and QIIME 2.

    “For instance, it is an integral component of many popular open-source applications such as Galaxy (Afgan et al. 2018), Nextstrain (Hadfield et al. 2018), OrthoFinder (Emms and Kelly 2015), and QIIME 2 (Bolyen et al. 2019).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites Afgan et al. 2018 source

      Not checked in the free run.

    • cites Hadfield et al. 2018 source

      Not checked in the free run.

    • cites Emms and Kelly 2015

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    IQ-TREE 2 exploits partial terraces to improve tree search under partitioned models and achieves up to 4.5- and 8-fold speedups compared with IQ-TREE 1 and RAxML, respectively.

    “IQ-TREE 2 exploits partial terraces to improve tree search under partitioned models and achieves up to 4.5- and 8-fold speedups compared with IQ-TREE 1 and RAxML, respectively (Chernomor et al. 2016).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Chernomor et al. 2016 source

      Not checked in the free run.

  • May combine two claims.

    Interpretationcitation

    The high performance of IQ-TREE results from the efficient integration of novel phylogenetic methods that improve the three key steps in phylogenetic analysis: fast model selection via ModelFinder, an effective tree search algorithm, and a novel ultrafast bootstrap approximation.

    “The high performance of IQ-TREE results from the efficient integration of novel phylogenetic methods that improve the three key steps in phylogenetic analysis: fast model selection via ModelFinder (Kalyaanamoorthy et al. 2017), an effective tree search algorithm (Nguyen et al. 2015), and a novel ultrafast bootstrap approximation (Minh et al. 2013; Hoang et al. 2018).”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites Kalyaanamoorthy et al. 2017 source

      Not checked in the free run.

    • cites Nguyen et al. 2015 source

      Not checked in the free run.

    • cites Minh et al. 2013 source

      Not checked in the free run.

17 of 17

  • The paper relies on this without support.

    Missing data are almost unavoidable in phylogenomic data sets.

    “Missing data—when some loci or sites are absent for some species—are almost unavoidable in phylogenomic data sets.”
1 of 1

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • The maximum likelihood phylogeny was performed under 1000 Ultra-Fast bootstrap replications.

    “The maximum likelihood phylogeny was performed using IQTREE (IQ-TREE multicore version 1.6.12 for Mac OS X 64-bit) software [62], under 1000 Ultra-Fast bootstrap replications [63].”

    From Emergent and Neglected Equine Filariosis in Egypt: Species Diversity and Host Immune Response (2022) · cites this paper as [62]

    The Fast Topology Tests section (page unavailable) says IQ-TREE 2 provides bootstrap resampling schemes and cites work on the ultrafast bootstrap approximation. Our copy does not specify 1,000 ultrafast bootstrap replicates or document the citing authors’ analysis, so it supports the method but not that number or its use in this phylogeny. (LLM, conf=0.83)

    Experimental — accuracy not yet measured

  • In Wang et al., 6 out of 17 RBD-specific mAbs isolated from mRNA-1273 vaccinated individuals showed more than 100-fold neutralisation loss against N501Y mutant, consistent with the loss of neutralisation by 5 out of 29 RBM-specific mAbs described in this study.

    “In the study by Wang et al., 6 out 17 RDB-specific mAbs isolated from mRNA-1273 vaccinated individuals showed more than 100-fold neutralisation loss against N501Y mutant, a finding that is consistent with the loss of neutralisation by 5 out 29 RBM-specific mAbs described in this study.”

    From Sensitivity of SARS-CoV-2 B.1.1.7 to mRNA vaccine-elicited antibodies (2021) · cites this paper as [41]

  • Wang et al. observed a reduction in neutralization titres from mRNA-elicited antibodies in volunteers who received two doses (mRNA-1273 and BNT162b2) using pseudoviruses carrying the N501Y mutation.

    “A reduction in neutralization titres from mRNA-elicited antibodies in volunteers who received two doses (using both mRNA-1273 and BNT162b2 vaccines) was also observed by Wang et al.41 using pseudoviruses carrying the N501Y mutation.”

    From Sensitivity of SARS-CoV-2 B.1.1.7 to mRNA vaccine-elicited antibodies (2021) · cites this paper as [41]

Cites

Flagged first, then by citations

Research works printed but not linked: 5

  1. Felsenstein, 2004, Inferring phylogenies
  2. Grama, 2003, Introduction to parallel computing
  3. Lemey, 2009, The phylogenetic handbook: a practical approach to phylogenetic analysis and hypothesis testing, 10.1017/CBO9780511819049 · DOI
  4. Minh, 2018, 10.1101/487801 · DOI
  5. Snir, 1998, MPI: the complete reference—the MPI core

Cited by

Sorted by citations

Authors
Minh, B. Q., Schmidt, H. A., Chernomor, O., Schrempf, D., Woodhams, M. D., von Haeseler, A., & Lanfear, R.
Affiliations
  • Research School of Computer Science, Australian National University, Canberra, ACT, Australia
  • University of Vienna (AT)
  • Eötvös Loránd University (HU)
  • University of Tasmania (AU)
  • Center for Integrative Bioinformatics Vienna, Max Perutz Labs, University of Vienna and Medical University of Vienna, Vienna, Austria
  • Australian National University (AU)
ORCID
Venue
Molecular Biology and Evolution
Publisher
Oxford University Press (OUP)
Published
2020-02-03
Volume
37
Issue
5
Pages
1530-1534
ISSN
  • 0737-4038 (ISSN-L)
  • 1537-1719
DOI
10.1093/molbev/msaa015
PMC
PMC7182206
OpenAlex
W3005235420
Semantic Scholar
3c9016ea7502857551053972659528d229aeb800
Cited by
16,873
References
49
Citations by year
2026: 4136 · 2025: 4208 · 2024: 3276 · 2023: 2242 · 2022: 1757 · 2021: 1019 · 2020: 222 · 2019: 1 · 2017: 1 · 2016: 1 · 2013: 1
FWCI
430.88
Type
journal article
Language
en
Open access
Licence
Locations
  • Molecular Biology and Evolution (publishedVersion, cc-by)
  • bioRxiv (Cold Spring Harbor Laboratory) (acceptedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • ANU Open Research (Australian National University) (submittedVersion, cc-by)
Funders
  • Austrian Science Fund (I-2805-B29)
  • Australian National University Futures Scheme
  • European Research Council
  • European Union's Horizon 2020 research and innovation programme (714774)
Topics
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › Genomics and Phylogenetic Studies
  • Life Sciences › Biochemistry, Genetics and Molecular Biology › Molecular Biology › RNA and protein synthesis mechanisms
  • Life Sciences › Agricultural and Biological Sciences › Ecology, Evolution, Behavior and Systematics › Plant Diversity and Evolution
Keywords
IQ-TREE, phylogenetic inference, maximum likelihood estimation, sequence evolution models, genomic data, phylogenomics
MeSH terms
Models, Genetic, Phylogeny, Software, Evolution, Molecular, Genomics

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review