Verbalizing phylogenomic conflict: Representation of node congruence across competing reconstructions of the neoavian explosion

Bioinformatics

Syntactic and semantic conventions

1. Taxa are models, concepts are mimics. We typically refrain from using the terms “taxon”, “taxa”, or “clade(s)”. We take taxa to constitute evolutionary, causally sustained entities whose members are manifested in the natural realm. The task for systematics is to successively approximate the identities and limits of these entities. Thus, we assign the status of ‘models’ to taxa, which systematists aim to ‘mimic’ through empirical theory making. This perspective allows for realism about taxa, and also for the possibility to let our representations stand for taxa [31], at any given time and however imperfectly, to support evolutionary inferences.

In reserving a model status for taxa, we can create a separate design space for the human theory- and language-making domain. In the latter, we speak only of taxonomic or phylogenomic concepts–the products of inference making [21].

2. Sameness is limited to the same source. Therefore, for the purpose of aligning the neoavian explosion use case, we need not speak of the “same taxa” or “same clades” at all. Similarly, we need not judge whether one reconstruction or the other more closely aligns with deep-branching avian taxa, i.e., which is (more) ‘right’? Instead, our alignment is only concerned with modeling congruence and conflict across two sets of concept hierarchies. The concepts are labeled with the “sec.” convention to maintain a one-to-one modeling relationship between concept labels and concepts (clade identity theories). Accordingly, there is also no need to say that, in recognizing each a concept with the taxonomic name Neornithes, the two author teams are authoring “the same concept”. Instead, we model the two labels 2015.Neornithes and 2014.Neornithes, each of which symbolizes an individually generated phylogenomic theory region. As an outcome of our alignment, we may say that these two concepts are congruent, or not, reflecting the intensional alignment (to be specified below) of two phylogenomic theories. But, by virtue of their differential sources (authorship provenance), the two concepts 2015.Neornithes and 2014.Neornithes are never “the same”. “Sameness” is limited in our approach to concepts whose labels contain an identical taxonomic name and which originate from a single phylogenomic hierarchy and source. That is, 2015.Neornithes and 2015.Neornithes are (labels for) the same concept.

Knowledge representation and reasoning

The methods used herein are consistent with [14, 26, 32]. They utilize three core conventions: (1) taxonomic concept labels to identify concepts; (2) is_a relationships to assemble single-source hierarchies via parent/child relationships; and (3) RCC–5 articulations to express the relative congruence of concept regions across multi-sourced hierarchies. The RCC–5 articulation vocabulary entails (with corresponding symbol): congruence (= =), proper inclusion (>), inverse proper inclusion (<), overlap (><), and exclusion (!). Disjunctions of these articulations are a means to express uncertainty; as in: 2015.Neornithes {= = or > or <} 2014.Neornithes. All possible disjunctions generate a lattice of 32 relationships (R32), where the “base five” are the most logically constraining subset [33].

The alignments are generated with the open source Euler/X software toolkit [28]. The toolkit ingests multiple trees (T1, T2, T3, etc.) and articulation sets (A1–2, A2–3, etc.), converting them into a set of logic constraints. Together with other default or facultative constraints (C) needed for modeling tree hierarchies, these input constraints are then submitted to a logic reasoner that provide two main services. First, the reasoner infers whether all input constraints are jointly logically consistent, i.e., whether they permit at least one “possible world”. Second, if consistency is attained, the reasoner infers the set of Maximally Informative Relations (MIR). The MIR constitute that unique set of RCC–5 articulations for every possible concept pair across the input sources from which the truth or falseness of any relationship in the R32 lattice can be deduced [14, 26, 33]. Many toolkit options and functions are designed to encode variable alignment input and output conditions, and to interactively obtain adequately constrained alignments. The toolkit also features a stylesheet-driven alignment input/output visualization service that utilizes directed acyclical graphs [28]. A step-wise account of the user/toolkit workflow interaction is provided in [26].

Special challenges for multi-phylogeny alignments

Aligning phylogenomic trees entails several special representation and reasoning challenges. We address three aspects here that have not been dealt with extensively in previous publications.

1. Representing intensional parent concept congruence via locally relaxed coverage. The first challenge relates directly to the issue of parent node identity. Unlike comprehensive classifications or revisions [14, 26, 34], phylogenomic reconstructions typically do not aspire to sample low-level entities exhaustively. Instead, select exemplars are sampled among all possible low-level entities. The aim is to represent lower-lever diversity sufficiently well to infer reliable higher-level relationships. Often, terminal sampling is not only incomplete for any single reconstruction, but purposefully complementary to that of other analyses. Generating informative genome-level data remains resource-intensive [10]. This makes it prudent to coordinate terminal sampling globally, by prioritizing the reduction of gaps over redundant terminal sampling. In the case of 2015.PEA (198 terminals) versus 2014.JEA (48 terminals), only 12 species-level concept pairs have labels with identical taxonomic names.

By default, the logic toolkit applies a coverage constraint to every input concept region. Coverage means that the region of a parent is strictly circumscribed by the union of its children [35]. However, this constraint is relaxable, either globally for all concepts, or locally for select concepts. To relax coverage locally, the prefix “nc_” (no coverage) is used in the input, as in 2014.nc_Psittacidae. This means: either a parent concept’s referential extension is circumscribed by the union of its explicitly included children, or there is a possibility of additional children being subsumed under that parent but not mentioned in the source phylogeny. Either scenario can yield consistent alignments. In other words, if a parent concept has relaxed coverage, it can attain congruence with another parent concept in spite of each parent having incongruent sets of child concepts.

Managing coverage in the toolkit input is not trivial. Relaxing coverage globally is akin to saying “anything goes”, i.e., any parent could potentially include any child. This would yield innumerable possible worlds, and therefore has no value for our purpose. On the other hand, applying coverage globally means–counter-intuitively in the case of phylogenomic trees–that only parents with completely congruent sets of children can themselves attain congruence. The challenge for experts providing the input is thus to relax coverage locally, and strictly in the service of ‘neutralizing’ lower-level sampling differences between trees that should not yield conflict at higher levels.

The effect of locally relaxed coverage is illustrated in Figs 14, using the example of parrots– 2015./2014.Psittaciformes. At the species level, the author teams sampled wholly exclusive sets of concepts for this alignment region (Figs 1 and 3). Even at the genus level, only 2015./2014.Nestor is redundantly sampled, yet with the articulation: 2015.Nestor_meridionalis ! 2014.Nestor_notabilis at the child level. Therefore, if no species-level concept sec. 2015.PEA has an explicitly sampled and congruent region in 2014.Psittaciformes, and, vice-versa, no species-level concept sec. 2014.JEA has such a region in 2015.Psittaciformes, then under global application of the coverage constraint we obtain the alignment: 2015.Psittaciformes ! 2014.Psittaciformes (Fig 2). The absence of even partial concept region overlap at the terminal level ‘propagates up’ to the highest-level parent concepts, which are therefore also exclusive of each other.

thumbnail

Fig 1. Input visualization for the 2015./2014.Psittaciformes alignment, with coverage globally applied.

In all toolkit visualizations, the input and aligned, non-congruent concepts sec. 2015.PEA are shown as green rectangles (T2−18 concepts). Input and aligned, non-congruent concepts sec. 2014.JEA are shown as yellow octagons (T1−6 concepts). Congruent sets of aligned, multi-sourced concepts (first shown in Fig 4) are rendered in gray rectangles with rounded corners. In this input visualization, each phylogenomic tree is separately assembled via parent/child (is_a) relationships (solid black arrows). All species-level concepts sec. 2015.PEA and 2014.JEA are exclusive of each other. Under strict application of the coverage constraint, this is represented by asserting eight articulations (dashed magenta arrows) of disjointness (!) of each species-level concept from the other-sourced order-level concept. The legend indicates the numbers of nodes and edges for each input tree, parent/child relationships, and expert-asserted input articulations. See also S1 File.


https://doi.org/10.1371/journal.pcbi.1006493.g001

thumbnail

Fig 2. Alignment visualization for the 2015./2014.Psittaciformes alignment, with coverage globally applied.

This alignment corresponds to the Fig 1 input, and shows reasoner-inferred non-/congruent concepts and articulations (see legend)–i.e., none in this particular case. The reasoner infers 108 logically implied articulations that constitute the set of MIR. See also S2 File. Although the input and alignment of Figs 1 and 2 are empirically defensible, they fail to capture certain intuitions we have regarding the higher-level 2015./2014.Psittaciformes relationship. For instance, we may wish to say: “Sure, the author teams sampled complementary species-level concepts. Yet these trees are not actually in conflict. At higher levels, there likely is agreement that parrots are parrots, and non-parrots are non-parrots”. That is: 2015.Psittaciformes = = 2014.Psittaciformes. To obtain this intuitive alignment, we have to locally relax coverage at select lower levels (Fig 3). In particular, 2015.PEA include five genus- and species-level concepts under 2015.Psittacidae that have no corresponding region under 2014.Psittacidae. However, if we relax coverage for 2014.Psittacidae–i.e., we assert 2014.nc_Psittacidae as an input constraint–then we can include each of these; for instance: 2015.Probosciger_aterrimus < 2014.Psittacidae, 2015.Psittacus_erithacus < 2014.Psittacidae, etc. Conversely, if we locally relax coverage for 2015.Psittacidae (2015.nc_Psittacidae), we can specify 2014.Melopsittacus_undulatus < 2015.Psittacidae. At the genus level, we can align 2015.Nestor = = 2014.Nestor if we relax coverage for each (2015.nc_Nestor, 2014.nc_Nestor), in spite of the mutually exclusive species-level concepts sampled. Jointly, these four instances of relaxing coverage render the articulation 2015.Psittacidae = = 2014.Psittacidae consistent, and hence also 2015.Psittaciformes = = 2014.Psittaciformes (Fig 4).


https://doi.org/10.1371/journal.pcbi.1006493.g002

thumbnail

Fig 3. Input visualization for the 2015./2014.Psittaciformes alignment, with coverage locally relaxed.

Compare with Fig 1. Here, coverage is relaxed for two family-level concepts (2015./2014.nc_Pittacidae) and two genus-level concepts (2015./2014.nc_Nestor). The eight species-level concepts of the alignment are correspondingly included as members of these higher-level concepts. In addition, three instances of congruence are asserted for 2015./2014.{Psittaciformes, Psittacidae, Nestor}. See also S3 File.


https://doi.org/10.1371/journal.pcbi.1006493.g003

thumbnail

Fig 4. Alignment visualization for the 2015./2014.Psittaciformes alignment, with coverage locally relaxed.

Compare with Fig 2. Local relaxing of coverage, and assertions of congruence of paired higher-level concepts (Fig 3), will yield the intuitive alignment of 2015.Psittaciformes = = 2014.Psittaciformes, 2015.Psittacidae = = 2014.Psittacidae, and 2015.Nestor = 2014.Nestor; in spite of wholly incongruent sampling of species-level concepts. The reasoner infers 160 logically implied articulations that constitute the set of MIR. See also S4 File.


https://doi.org/10.1371/journal.pcbi.1006493.g004

Asserting higher-level node congruence in light of lower-level node incongruence requires a conception of node identity that affirms counter-factual statements of the following type: if 2014.JEA had sampled 2014.Psittacus_erithacus, then the authors would have included this species-level concept as a child of 2014.Psittacidae. This is to say that 2015./2014.Psittacidae, and hence their respective parents, are intensionally defined [25, 36, 37].

Using a combination of published topological information (and support), more or less direct reiterations of phenotypic traits (cf. discussions and supplementary data of 2015.PEA and 2014.JEA), and trained judgment [30], we align these concept regions as if there are congruent property criteria that each region entails, i.e., something akin to an implicit set of synapomorphies or uniquely diagnostic features. Of course, the phylogenomic data provided by 2015.PEA and 2014.JEA do not signal intensional definitions directly. But neither do their genome-based topologies for parrots provide evidence to challenge the status of such definitions as previously proposed [38]. In addition, particularly 2015.PEA (supplementary information; sections on “detailed justification for fossil calibrations” and “detailed phylogenetic discussion; pp. 3–21) provide a provide an in-depth account of how their preferred topology relates to published, property-centered circumscriptions of dozens of higher-level clade concepts. We have to assume, fallibly and non-trivially, that such topology-to-synapomorphy relations are also implied by JEA.2014, as reflected (inter alia) in their discussion.

Three clarifications are in order. First, Region Connection Calculus is at best a means of translating the signal of an intensional definition. The congruent (= =) symbol means, only: two regions are congruent in their extension. The RCC–5 vocabulary is obviously not appropriate for reasoning directly over genomic or phenomic property statements. The reasoner does not assess whether 2015.Psittacidae, or any included child or aligned concept, has ‘the relevant synapomorphies’. Doing so would not be trivial even if property-based definitions were provided for all higher-level node concepts, because we would still have to make theory-laden assumptions about their congruent phylogenomic scopes [26, 39, 40]. Second, we are not providing detailed textual narratives that would justify each assertion of higher-level congruence. Such narratives are possible, and even needed to understand disagreements, because they explain the reasoning process behind an expert-made assertion. However, our main objective here is to focus on the issue of RCC–5 translation of systematic signals; not on a character-by-character dissection of each congruent articulation. Third, a sensible intensional alignment strategy uses a minimal number of instances of locally relaxed coverage in order to compensate for differential child sampling at lower levels, so that parent coverage can remain in place at higher levels to expose incongruent node concepts. The benefits of this strategy will be shown below.

2. Representing clade concept labels. Our modeling approach requires that every region in each source tree receives a taxonomic or clade concept label. However, the source publications only provide such labels for a subset of the inferred nodes. In particular, 2015.PEA (p. 570: Fig 1) obtained 41 nodes above the ordinal level. Of these, 17 nodes (41.5%) were explicitly labeled in either the published figure or supplement (pp. 9–12). The authors also cite [20] as the primary source for valid name usages, yet that list is not concerned with supra-ordinal names. Similarly, 2014.JEA (p. 1322: Fig 1) inferred 37 nodes above the ordinal level, of which 23 nodes (62.2%) were given an explicit label. They provide an account (cf. supplementary materials SM6: 22–24) of their preferred name usages, sourced mainly to [20] and [41].

In assigning clade concept labels at the supra-ordinal level when the authors may have failed to do so (consistently), we nevertheless made a good faith effort–through examination of the supplementary information and additional sources [1, 3, 42, 43, 44, 45, 46, 47]–to represent the authors’ preferred name usages. Where usages were not explicit, we selected the only or most commonly applied clade concept name at the time of publication. This effort yielded 13 additional labels for 2015.PEA (Table 1), and 7 such labels for 2014.JEA (Table 2).

thumbnail

Table 1. Supra-ordinal clade concept labels used for the phylogenomic tree of 2015.PEA, with sources from which the names were obtained.

“Franz et al. 2018” means: the label was assigned pragmatically in this study. See main text for further detail.


https://doi.org/10.1371/journal.pcbi.1006493.t001

thumbnail

Table 2. Supra-ordinal clade concept labels used for the phylogenomic tree of 2014.JEA, with sources from which the names were obtained.

“Franz et al. 2018” means: the label was assigned pragmatically in this study. See main text for further detail.


https://doi.org/10.1371/journal.pcbi.1006493.t002

If no suitable label was available, we chose a simple naming convention of adding “_Clade1”, “_Clade2”, etc., to the available and immediately higher-level node label, e.g. 2014.Passerea_Clade1. The numbering of such labels along the tree topology starts with the most immediate child of a properly named parent, and typically follows down one section of the source tree entirely (“depth-first”), before continuing with the higher-level sister section. Using this approach, we added 11 labels for 2015.PEA (Table 1) and 7 labels for JEA.2014 (Table 2). If greater numbers of labels need to be generated, including siblings, then it is sensible to have a rule for ordering sibling nodes, e.g. by assigning the next-lowest number to the sibling whose child’s name appears first in the alphabet. Our numbering of the labels 2014.Passerea_Clade2 (child with first-appearing letter: 2014.Ardeae) and 2014.Passerea_Clade3 (child: 2014.Cursorimorphae) adhere to this rule.

The clade concept labeling convention was not applied below the family level, where instead phylogenomic resolution was collapsed into polytomy (exception: Figs 14). In the case of 2014.JEA, only four family-level concepts include two children, whereas the remainder have a single child sampled. Resolving the monophyly of subfamilial clade concepts was not the primary aim of 2014.JEA. The same applies to 2015.PEA, who sampled 104/125 family-level concepts with only 1–2 children.

3. Representing phylogeny/classification paraphyly. A third, relatively minor challenge is the occurrence of clade concepts in 2015.PEA’s phylogenomic tree that are not congruently aligned with higher-level concepts of [20]. We highlight these instances here because they represent a widespread phenomenon in phylogenomics. It is useful to understand how such discrepancies can be modeled with RCC–5 alignments (Figs 5 and 6).

thumbnail
Fig 5. Input visualization of the alignment of the phylogenomic reconstruction of passeriform clade concepts sec. 2015.PEA–prefixed with “Phylo2015″–with the corresponding classification concepts sec. Gill & Donsker (2015) [20]–prefixed with “Class2015”.

The phylogenomic topology renders that of Class2015.Eurylaimidae paraphyletic, and hence the name “Eurylaimidae” is not represented in any clade concept label sec. 2015.PEA. See also Prum et al. (2015). See also S5 File.


https://doi.org/10.1371/journal.pcbi.1006493.g005

thumbnail
Fig 6. Alignment visualization corresponding to Fig 5.

The alignment shows an overlapping articulation (dashed blue line) between the phylogenomic clade concept sec. 2015.PEA (Phylo2015.Passeriformes_Clade1) and the Eurylaimidae sec. Gill & Donsker (2015) [20] (Class2015.Eurylaimidae). The two dashed red arrows symbolize reasoner-inferred relationships not explicit in the input constraints. See also S6 File.


https://doi.org/10.1371/journal.pcbi.1006493.g006

Fig 5 exemplifies the phylogenetic tree/classification incongruence observed in 2015.PEA. The authors state (supplementary Table 1, p. 1): “Taxonomy follows Gill and Donsker (2015; fifth ed)”. As shown in Fig 5, their phylogeny accommodates four sampled genus-level concepts that would correspond to children of the family-level concept Eurylaimidae sec. Gill & Donsker (2015) [20]. However, these concepts are arranged paraphyletically in relation to the reference classification. There is no parent concept that can be labeled 2015.Eurylaimidae and would not also (1) include 2015.Pittidae, i.e., 2015.Passeriformes_Clade1 in Fig 6, or (2) just represent aligned subset of the Eurylaimidae sec. Gill and Donsker (2015) [20], i.e., 2015.Passeriformes_Clade2 or 2015.Passeriformes_Clade3 in Fig 6. The concept Eurylaimidae sec. Gill and Donsker (2015) [20] has an overlapping (><) articulation with 2015. Passeriformes_Clade1.

In summary, our approach represents non-monophyly as an incongruent alignment of the phylogenomic tree and the source classification used to provide labels for that tree’s monophyletic clade concepts. There are four distinct regions in the phylogeny of 2015.PEA where such alignments are needed: {Caprimulgiformes, Eurylaimidae, Hydrobatidae, Procellariidae, Tityridae} sec. Gill & Donsker (2015) [20]. Each of these is provided in the S7S9 Files.

Configuration of input constraints and alignment partitioning

The source phylogenies specify 703 and 216 clade or taxonomic concepts, respectively. The frequent instances of locally relaxed coverage increase the reasoning complexity in relation to multi-classification alignments [14], making specialized RCC–5 reasoning useful [48]. The reasoning and visualization challenges commend a partitioned alignment approach. To keep the Results concise, we show visualizations of the larger input and alignment partitions only in the Supporting Information. A detailed account of the input configuration and partitioning workflow is given below.

Underlying all alignments is the presumption that at the terminal (species) level, the taxonomic concept labels of 2015.PEA and 2014.JEA are reliable indicators of either pairwise congruence or exclusion [14, 26, 32]. That is, e.g., 2015.Cariama_cristata = = 2014.Cariama_cristata, or 2015.Charadrius_hiaticula ! 2014.Charadrius_vociferus. Because the time interval separating the two publications is short in comparison to the time needed for taxonomic revisions to effect changes in classificatory practice, the genus- or species-level taxonomic concepts are unlikely to show much incongruence; though see [49] or [50]. We note that 2015.PEA (p. 571) use the label 2015.Urocolius(_indicus) in their phylogenomic tree, which also corresponds to the genus-level name endorsed in [20] Gill & Donsker (2015). However, in their Supplementary Table 1 the authors use 2015.Colius_indicus. We chose 2015.Urocolius and 2015.Urocolius_indicus as the labels to apply in the alignments.

The toolkit workflow favors a partitioned, bottom-up approach [29]. The process of generating, checking, and regenerating input files must be handled ‘manually’ on the desktop (note: improved workflow documentation and semi-automation of input-output-input changes are highly desirable). The performance of different toolkit reasoners was benchmarked in [28].

To work efficiently, the large problem of aligning all concepts at once is broken down into multiple smaller alignment problems, e.g. 2015./2014.Psittaciformes (Figs 3 and 4). To manage one particular order-level alignment, we start with assembling each input phylogeny separately, with relaxed coverage applied as needed (Fig 3). The RCC–5 articulations for low-level concept pairs are provided incrementally, e.g., in sets of 1–5 articulations at a time. Following such an increment, the toolkit reasoning process is re-/deployed to validate input consistency and infer the number of possible worlds. There is an option to specify that only one possible world is sought as output, which is equivalent to just checking for input consistency, as opposed to inferring all possible worlds. Doing so saves time as long as the input remains (vastly) under-specified. The stepwise approach of adding a small number of articulations at a time leads to increasingly constrained alignments, while minimizing the risk of introducing many new. difficult-to-diagnose inconsistencies.

Once a set of small, topographically adjacent alignment partitions is well specified, these can serve as building blocks for the next, larger partition. Hence, the basic sequence of building up larger alignments is: (1) obtain a well-specified low- (order- or family-) level alignment; (2) record the inferred parent-level articulations from this alignment; (3) propagate the latter–now as low-level input articulations–for the next, more inclusive alignment; (4) as needed, prune the lowest-level (sub-ordinal) input concepts and articulations of (1) from this alignment; (5) repeat (1) to (4) for another paired region; (6) assemble the more inclusive alignment by (manually) connecting the pruned, propagated concepts and articulations from two or more lower-level alignments, by adding to them the higher-level concepts from each input phylogeny. Depending on the interplay between (ranked) higher-level names recognized in each phylogeny and the number of terminal concepts sampled, steps (1) to (6) may be iterated once (e.g., 2015./2014.{Falconiformes, Psittaciformes}) or multiple times (e.g., 2015./2014.Passeriformes) to cover a supra-/ordinal alignment. An example of the latter is the 2015./2014.Passerimorphae alignment, which includes two order-level concepts and their children in each source phylogeny. Such mid-level partitions eventually form the basis for the largest alignment partitions, e.g. 2015./2014.Telluraves.

Sometimes, coverage will have to be relaxed even at higher levels. In all, 2014.JEA sample children of 34 order-level concepts in their phylogeny, whereas 2015.PEA recognize 40 order-level concepts. The latter authors represent four order-level concepts for which no analogous children are included in 2014.JEA, i.e.: 2015.{Apterygiformes, Casuariiformes, Ciconiiformes, Rheiformes}. Three of these are assigned to 2015.Palaeognathae, whereas 2015.Ciconiiformes are subsumed under 2015.Pelecanimorphae–in each case under relaxed parent coverage. The remaining 36 order-level concepts sec. 2015.PEA show some child-level overlap with those of 2014.JEA.

Our partitioning approach for this use case started with specifying the input constraints for nearly 35 paired order-level concepts and their respective children, as demonstrated in Figs 3 and 4. The largest order-level partition is 2015./2014.Passeriformes, with 148 x 22 input concepts, seven instances of relaxed parent coverage, and 101 input articulations. This alignment completes in less than 15 seconds on an individual 2.0 GHz processor, yielding 3,256 MIR.

As the partitions grew, we configured the following six, non-overlapping alignments as building blocks for the global alignment: 2015./2014.Palaeognathae (34 x 12 input concepts, four instances of relaxed coverage, and 25 articulations; same data sequence used for following alignments), 2015./2014.Galloanserae (49, 16, 7, 46), 2015.Columbaves/2014.Columbimorphae + 2014.Otidimorphae (53, 37, 13, 37), 2015.Strisores/2014.Caprimulgimorphae (44, 17, 8, 32), 2015./2014.Ardeae (100, 55, 19, 75), and the largest partition of 2015./2014.Telluraves (316, 104, 37, 241).

At the next more inclusive level, the inferred congruence of 2015.Telluraves = = 2014.Telluraves presented an opportunity to partition the entire alignment into two similarly sized regions, where the complementary region includes all 2015./2014.Neornithes concepts (392, 174, 58, 259), except those subsumed under 2015./2014.Telluraves, which are therein only represented with two concepts labels and one congruent articulation. These two complements are the core partitions that inform our use case alignment, globally. The corresponding S10 and S11 Files include the input constraint (.txt) and visualization (.pdf) files, along with the alignment visualization (.pdf) and MIR (.csv).

The two large partitions yield unambiguous RCC–5 articulations from the species concept level to that of 2015./2014.Neornithes. They can be aggregated into a synthetic, root-to-order level alignment, where all subordinal concepts and articulations are secondarily pruned away (see above). Such an alignment retains the logic signal derived from the bottom-up approach, but represents only congruent order-level concept labels as terminal regions, except in cases where there is incongruence. We present this alignment as an analogue to Fig 1 in [4] (p. 515), and compare how each conveys information about congruent and conflicting higher-level clade concepts.

Lastly, we further reduce the root-to-order alignment to display only 5–6 clade concept levels below the congruent 2015./2014.Neoaves. This region of the alignment is the most conflicting, and therefore forms the basis for our Discussion.

Articles You May Like

Spider Uses Web As Slingshot To Ensnare Prey, Scientists Find
The Return On Investment Of Healthcare Compliance
Species facing climate change could find help in odd place: Urban environments
Polymers jump through hoops on pathway to sustainable materials
Stopping Key Tech Exports To China Could Backfire, Researchers And Firms Say

Leave a Reply

Your email address will not be published. Required fields are marked *