Chapter Four · failure evidence
What Phylogenetic Tree Reconstruction got wrong, from 42 dissertations
The records evaluate various computational, algorithmic, and data preparation strategies for phylogenetic tree inference across diverse biological and linguistic datasets. Many attempts failed or were rejected due to insufficient phylogenetic signal from restricted markers, reticulate evolutionary events violating tree structures, matrix filtering artifacts, and inaccurate distance approximations. These records come from PhD theses at 16 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.
Single markers and isolated partitions lack sufficient phylogenetic signal and create erroneous groupings
Relying on individual genes, single morphological features, or restricted sequence regions produces unresolved polytomies and weak branch support. These narrow data sources are often distorted by recombination, horizontal gene transfer, and homoplasy compared to whole genomes or multi-locus transcriptomic datasets.
Tried and failed
single-locus phylogenetic tree reconstruction applied to species tree inference in amphibians. Outcome: no signal. Reason: insufficient phylogenetic resolution leading to discordance and low posterior support
Adaptation during range expansion: a phylogenetic, population genetic, and physiological perspective · Cornell
Considered and rejected
Considered and rejected: Rejected individual nuclear gene trees as standalone phylogenetic markers due to polytomies and weak nodal support.
Exome analyses and species delimitations of the Heteromys pictus-spectabilis species complex · Texas Tech
Tried and failed
phylogenetic reconstruction using internal coding sequences applied to endogenous retrovirus subfamily classification. Outcome: no signal. Reason: coding sequences provided insufficient granularity to resolve subfamilies compared to non-coding terminal repeats
RECOMBINATION-MEDIATED REGULATORY EVOLUTION OF HUMAN ENDOGENOUS RETROVIRUS HERVH · Cornell
Tried and failed
phylogenetic inference from single anatomical partition applied to morphological evolutionary tree reconstruction. Outcome: worse than baseline. Reason: isolated postcranial characters lacked sufficient phylogenetic signal and introduced severe homoplasy-driven misplacements
Effects of Additional Postcranial Data on the Topologies of Early Eocene Primate Phylogenies · UT Austin
Tried and failed
concatenating top marker genes for phylogenetic distance applied to microbial eukaryotic genome taxonomic classification. Reason: diminishing returns with plateauing resolution unable to match whole-genome sequence resolution
Expanding The Bioinformatics Toolbox for Diversity and Taxonomic Studies of Microbial Eukaryotic Pathogens · Georgia Tech
Considered and rejected
Considered and rejected: Rejected using mitochondrial genomes for deep heterobranch phylogenetic nodes because they perform poorly compared to transcriptomic datasets.
Tried and failed
single-gene marker phylogenetic classification applied to bacterial strain taxonomic identification. Outcome: did not generalise. Reason: horizontal gene transfer and recombination caused single-marker phylogeny to produce non-monophyletic groups
Bacterial Plant Pathogen Identification using Genomics and Metagenomics · Virginia Tech
Considered and rejected
Considered and rejected: Rejected classifying higher bacterial ranks and lineage divergences using 16S rRNA gene fragments alone due to insufficient phylogenetic resolution and primer bias, opting instead for genome-wide relative evolutionary divergence (RED).
Tried and failed
phylogenetic classification using single isolated morphological markers applied to dialect subgrouping and classification. Reason: progressive aspect markers were conflated with indicative markers, leading to erroneous lineage groupings
A Diachronic Analysis of Western Armenian Verbal Morphology · Penn
Strictly bifurcating tree models fail when evolutionary processes involve reticulation, introgression, or lateral transfer
Modeling admixed populations, hybridizing species, or mobile genetic elements on simple branching trees violates core evolutionary assumptions. These reticulate scenarios collapse tree resolution and necessitate explicit phylogenetic network or neighbor-net approaches.
Tried and failed
unadmixed phylogenetic tree modeling applied to admixed population placement. Reason: the lineage violates bifurcating tree assumptions due to being an admixed allelic sink
Establishing a Genetic System for Studying Cannabis Domestication · Harvard
Tried and failed
strictly bifurcating phylogenetic tree inference applied to hybridising and introgressing lineages. Outcome: no signal. Reason: reticulate evolution and extensive introgression violate strict tree-like branching assumptions
Tried and failed
phylogenetic stemmatology and genealogical classification applied to historical manuscript variant networks. Outcome: did not generalise. Reason: hybrid or contaminated textual variants prevented placement into standard tree-based lineages
Das Buch von geistlicher Armut: Überlieferung, Armutskonzept, Rezeption · Harvard
Tried and failed
bifurcating tree phylogenetic reconstruction applied to lineages with rampant horizontal gene flow. Outcome: no signal. Reason: widespread introgression breaks the fundamental assumption of a strictly branching evolutionary history
Considered and rejected
Considered and rejected: Rejected standard haplotypic ancestral state reconstruction on bifurcating trees due to pervasive historical gene flow and incomplete lineage sorting.
The genomic basis of repeated adaptation in deer mice · Harvard
Considered and rejected
Considered and rejected: Rejected standard bifurcating phylogenetic trees alone, adopting phylogenetic network analyses (SplitsTree Neighbor-Net) due to reticulate relationships and subgenome exchanges
Genotypes to Phenotypes: Genomic attributes that impact Genome-wide Associations in peanut (Arachis hypogaea. L). · Iowa State
Considered and rejected
Considered and rejected: Rejected dendrogram-based phylogenetic tree methods (e.g., ViPTree, GRAViTy) for classifying prophages because they flatten evolutionary relationships into a single hierarchy and cannot represent horizontal gene transfer and genome mosaicism.
Anti-phage defense as a driver of molecular innovation · MIT
Tried and failed
strict phylogenetic cladistic tree modeling applied to dialect evolution with lateral transfer. Outcome: did not generalise. Reason: Strict hierarchical trees cannot capture lateral transfer, reticulation, or horizontal contact dynamics.
A Diachronic Analysis of Western Armenian Verbal Morphology · Penn
Aggressive taxon pruning and character filtering destabilize consensus trees and discard informative phylogenetic signal
Removing ambiguous taxa or suspected homoplastic characters frequently eliminates essential phylogenetic information and causes severe consensus tree instability. Similarly, strict filtering of missing data reduces informative sites and creates topological conflict across clades.
Tried and failed
maximum parsimony phylogenetic consensus analysis applied to morphological character matrix with missing data. Outcome: unstable. Reason: fragmentary and unstable rogue taxa collapsed resolution across all internal clades in the consensus tree
Tried and failed
aggressive taxon pruning to resolve consensus polytomies applied to morphological maximum parsimony phylogenetic analysis. Outcome: unstable. Reason: removes highly complete, informative taxa and causes severe tree consensus instability
Tried and failed
filtering noisy features prior to parsimony inference applied to phylogenetic matrix resolution across diverse taxa. Outcome: data insufficient. Reason: removing suspected homoplastic characters severely reduced information content, preventing relationship resolution among taxa
Considered and rejected
Considered and rejected: Excluded Poxviridae from definitive replisome/transcription phylogenies due to unstable phylogenetic placement across diagnostic trees
Comparative and Evolutionary genomics of Nucleocytoviricota · Virginia Tech
Considered and rejected
Considered and rejected: Eliminated phylogenetic loci that produced more than 10 paralogs or displayed paralog trees without clear structure (unresolved homology).
Tried and failed
strict missing data filtering of loci applied to phylogenetic alignment matrix. Outcome: worse than baseline. Reason: filtering reduced total informative sites, decreasing branch support and creating topological conflict
The evolutionary history and diversity of the Mickey Mouse Plants (the genus Ochna) · Imperial
Considered and rejected
Considered and rejected: Evaluating single global phylogenetic trees across all taxa, discarded in favor of partitioned subset analyses because distant taxa artificially reduced informative site weights via homoplasy
Fast heuristic algorithms and parsimony criteria yield suboptimal tree likelihoods and unresolved clades
Heuristic tree search tools like FastTree2 produce likelihood scores that fall substantially short of robust probabilistic frameworks like IQ-TREE 2 and RAxML. In addition, maximum parsimony models often fail to resolve character partitions and yield less resolved topologies than probabilistic alternatives.
Tried and failed
maximum parsimony phylogenetic reconstruction across morphological partitions applied to fossil primate morphological dataset. Outcome: worse than baseline. Reason: produced a less resolved consensus tree and misclassified target taxon compared to baseline PAUP implementation
Effects of Additional Postcranial Data on the Topologies of Early Eocene Primate Phylogenies · UT Austin
Considered and rejected
Considered and rejected: FastTree2 was rejected for comprehensive phylogenetic inference and cross-comparisons because its likelihood scores were substantially worse than IQ-TREE 2, RAxML, and MrBayes.
Discovery and characterization of diverse microbial RNA-guided systems · MIT
Considered and rejected
Considered and rejected: Rejected FastTree2 for comprehensive phylogenetic cross-comparisons due to substantially worse likelihood scores than IQ-TREE 2, RAxML, or MrBayes.
Considered and rejected
Considered and rejected: Rejected parsimony in favor of maximum likelihood for phylogenetic reconstruction because ML provides a probabilistic model rather than minimal change steps.
Considered and rejected
Considered and rejected: Bayesian inference and Maximum Likelihood were rejected for primary morphological phylogenetic analyses due to computational intractability and reliance on questionable prior substitution/rate models for morphological data.
Approximate distance metrics and k-mer sketching distort pairwise relationships and hierarchical topologies
Heuristic distance calculations and default MinHash sketching sizes fail to preserve true edit distances and cophenetic correlations. These approximations introduce intra-clade noise and place closely related lineages into incorrect topological arrangements.
Tried and failed
approximate sequence distance via k-mer sketching applied to phylogenetic graph edge construction. Outcome: worse than baseline. Reason: Inconsistent distance estimations produced incorrect parent-child relationships compared to exact edit distance
Graph-Based Computational Approaches for Modeling Viral Evolution · Virginia Tech
Tried and failed
approximate pairwise distance estimation for hierarchical clustering applied to genome-scale phylogenetic reconstruction. Outcome: worse than baseline. Reason: heuristic distance approximations failed to accurately preserve hierarchical topology and cophenetic correlation across ranks
Tackling the current limitations of bacterial taxonomy with genome-based classification and identification on a crowdsourcing Web service · Virginia Tech
Tried and failed
hyperbolic graph embedding of accessibility matrix applied to phylogenetic relationship reconstruction. Reason: hyperbolic embedding placed closely related clusters nearer to distant nodes than expected ground-truth topology
Geometric Methods for Quantitative Analysis of Romance Languages · Harvard
Tried and failed
MinHash sketching distance estimation applied to closely related fungal phylogenetic reconstruction. Outcome: worse than baseline. Reason: Default sketch size caused excessive intra-clade variation and misplaced phylogenetic topology compared to alignment methods
Computational Genomics to Characterize the Distribution, Diversity and Drug Resistance of Fungal Pathogens · Georgia Tech
Outgroup and midpoint rooting techniques are biased by long branches and rate heterogeneity
Distant outgroups and divergent homologous lineages introduce long-branch attraction artifacts and recombination signals that distort root placement. Midpoint rooting and minimal ancestor deviation also fail when large rate differences across lineages systematically skew the inferred root position.
Tried and failed
outgroup rooting for phylogenetic tree reconstruction applied to divergent homologous gene families. Reason: long-branch attraction artifacts caused inconsistent and unresolved root placements across gene trees
Global distribution and ancient evolution of far-red photosynthesis · Imperial
Tried and failed
outgroup rooting with distant lineages applied to bacterial phylogenetic tree inference. Outcome: unstable. Reason: recombination events on long branches distorted phylogenetic signals and root placement
Genomic insights into the host range and interspecies transmission dynamics of Staphylococcus aureus · Cambridge
Considered and rejected
Considered and rejected: Rejected minimal ancestor deviation (MAD) and midpoint rooting for the global MMP tree because wide differences in branch lengths between metazoan and prokaryotic lineages systematically biased roots toward prokaryotes
Bridging the gap between microbial and metazoan evolutionary history · MIT
Left open by the authors
Problems the authors named and did not get to.
Left open
Evaluate MSA Transformer representations against phylogenetic trees and patristic distances rather than simple Hamming distances across protein sequences. Blocker: None
Revealing and exploiting coevolution through protein language models · EPFL
Left open
Investigate phylogenetic relationships among Ambystoma species to clarify morphological evolution patterns. Blocker: Lacks specific target dataset, genetic/morphological markers, or concrete phylogenetic methodology
Left open
Perform a comprehensive co-evolutionary analysis of host-parasite phylogenies across all bee host families using resolved host trees. Blocker: Detailed host phylogenetic trees across all bee host families were not yet available
THE EVOLUTION OF BROOD PARASITISM IN BEES AND OTHER ANIMALS · Cornell
Left open
Develop cytonuclear coevolution detection methods that account for incomplete lineage sorting, hybridization, and gene duplication in phylogenomics. Blocker: No concrete mathematical or algorithmic formulation specified for incorporating ILS/hybridization into ERC frameworks
Cytonuclear coevolution investigations in papilionoid legumes and medicago (fabaceae) · UT Austin
Left open
Gather phylogenetic and evolutionary evidence on the original selective pressures and timeline of pain origination. Blocker: Task is an open-ended biological/evolutionary research direction lacking concrete methodology or clear data scope
Foundations of pain · UT Austin
Left open
Compare phylogenetic or fossil occurrence patterns with reconstructed river capture landscapes to empirically test geomorphic controls on speciation rates. Blocker: Lacks specific target taxa, geographic locations, and concrete datasets to compare against landscape models.
Dynamic river networks drive landscape change and biological evolution · MIT
Left open
Cluster bacterial host source classes using phylogenetic trees to reduce classification output space complexity. Blocker: None
Applications of Machine Learning in Source Attribution and Gene Function Prediction · Virginia Tech
Left open
Develop robust machine learning algorithms for graphical models such as Ising models or phylogenetic trees with malicious nodes. Blocker: Task is framed as a broad research direction without specific algorithms, performance targets, or detailed formulations
Statistical Inference in the Differential Privacy Model · Cornell
Left open
Extend the PhyNE MCMC framework to jointly estimate niche model parameters and phylogenetic tree topology and branch lengths. Blocker: None
Phylogenetic Niche Modeling · Virginia Tech
Left open
Implement hierarchical or multi-scale flow matching architectures in PhylaFlow to scale phylogenetic tree inference to datasets beyond 50 leaves. Blocker: None
Foundation Model Guided Flow Matching on Tree Manifold for Differentiable Phylogenetics · Harvard
Checking a claim in this area?
We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.