Following

And why do you think that a tree based on non-morphological data "will give evolutionary information"? I would rather question the evolutionary information based on a study neglecting morphological data!

Suppose you have a set of taxa, and they share the same character state for characters that you believe are homologous (they have the same origin). You will consider that it is more parsimonious, or more likely, that this character state evolved only once through the history of the group, and therefore these taxa share a common ancestor, rather than these character states having evolved several times independently.

For example, it is more parsimonious to assume that flowers only appeared once through the evolution of plants and all flowering plants share a common ancestor that is not shared by other groups of plants, than to assume that flowers appeared several times independently through the evolutionary history of plants. By inferring a maximum parsimony tree using such characters, you are aiming at reconstructing the evolutionary relationships between organisms, hence you are doing phylogeny.

Of course there are many other possibilities, such as convergent evolution, reversals etc., but if you use enough characters to find out the most parsimonious or the most likely scenario that explains the evolution of your group, then you can build a tree showing the relationships between your organisms (but any tree will only be a hypothesis).

This is valid for all kind of characters as long as they are homologous, whether morphological, phylogenetic or other. Reconstructing evolutionary relationships using nucleotide sequences is exactly the same process. If a number of taxa have the same homologous sequence ACGT , whereas the homologous positions for another taxon are AAAA, it will be more parsimonious, and probably more likely to assume that they share a common ancestor not shared by the last taxon, rather than assume they have all independently undergone the same three mutations.

@ Roman Bohdan Hołyński, Nicolas Magain. Thank you sirs. I am working on the genus Acacia which is split into seven genera and out of which I am working on two genera (Senegalia and Vachellia). Though the topology within the genus is same, one analysis show Senegalia as a basal group while another analysis show Vachellia as a basal group. Both the analysis were parsimony. I used Mesquite and PAUP* software for the analysis. How I should interpret this? How I would say that one analysis is better than other? Is my dataset (number of species) is less?

Anup: "How I would say that one analysis is better than other?" - sometimes there is no way and you must leave the question unanswered, but usually you can find some "external evidence", e.g. evaluating the properties (and consequently susceptibility to various kinds of errors, like long branch effect, inadequate taxon- or character-sampling, definitions of characters, weighting of transformations, &c. - in your case especially "suspect" seems to me the selection of out-group) of the respective programs, as well as compatibility of results with biogeographical patterns, ecological characteristics of the analysed taxa, eventually (if available) palaeontological data. But you should always be aware that any result, any interpretation, any phylogenetical tree is only a more or less probable hypothesis, not anything like 100% sure fact ["... facts ... are nothing more than highly corroborated hypotheses" - Eldredge & Cracraft 1980]!

Of course your trees are phylogenetic hypotheses, especially those made with parsimony. (UPGMA is far too sensitive to long-branch problems. I'm not familiar with Bray-Curtis cluster analysis, but probably all cluster analysis methods are like UPGMA in this respect.) Unless each character has a lot of states, 35 characters are far too few to resolve the relationships of 30 terminal taxa. If most characters have only two states, as is usual, I would not really trust an analysis with less than three times as many characters as terminal taxa.

David: I think your statement is too categorical - it depends upon the specificities of particular taxa and particular charactes, upon taxon- and character-sampling, and, last not least, upon the specificity of the used program: I do not sufficiently understand the "working peculiarities" of other programs, I know nothing about Bray-Curtis or Mesquite, but e.g. my MICSEQ (based on "sequential parsimony": Minimal Iteratively Corrected distances and SEQuential reconstruction of ancestors) seems to be largely "robust" against the ratio "taxon/character" (and also against e.g. long-branch attraction and some other problems). Unfotunately, the computer program for MICSEQ has not been (and, alas, probably will never be...) put into the sufficiently "user-friendly" form as to allow wider propagation and using not only by myself [I am the author of the conception and algorithm, but the "computerization" was done by my physicist friend (my competences in computer science is practically zero) who - due to the pressure of other duties - has practically withdrawn from the "project"].

That's interesting. Has the method behind MICSEQ been published anywhere?

Mesquite is primarily a program for editing matrices, manipulating trees and tracing character evolution on trees. It can search for trees (at least using parsimony), but it's way too slow for all but the smallest matrices. (If only because it runs on Java, which makes it platform-independent.)

David: The basic idea and general outline of algorithm of MICSEQ have been described in my paper linked to below - since that time (2001) some details have been modified to allow solutions of more complicated situations, but without confusing the outline presented there. On the other hand, the computer program, over-optimistically mentioned in Acknowledgements as almost ready, was since radically modified several times, but the currently available version - MICSEQ 5.2 - has unfortunately been "locked up" in the "user-unfriendly" form, practically usable to nobody except me...

Thank you very much sirs. This is very informative. @ Roman Bohdan Hołyński and David Marjanović: How it is useful to give 'weight' to the particular character? Will this help more in resolving the tree?

Another question: How to code quantitative character, if it is overlapping. e.g. number of leaflets. one species show 10-15 leaflets, another species show 12-18 leaflets. How to 'code' this character? Or just ignore such kind of characters for analysis?

Thank you

Anup: The "source" of both your questions is the currently so fashionable "dread of subjectivism", although a moment of reflexion suffices to realize that any opinion, any evaluation, any decision is always largely subjective, based on (also subjective...) assessment of probability! So is this also with weighting: so-called "unweighted" (preferred by "objectivists") chaacters are in fact as subjectively weighted as any other: equal weighting (attributing the weight of 1 to each character) is no more objective ( but in most cases less adequate!) than subjectively differentiated! Of course we usually cannot reliably decide whether, say, the differences in structure of antennae (thread-like in carabids, lamellate in scarabaeids, clavate in dermestids, &c.) are exactly 3, 7, or 13 times more phylogenetically informative (more likely to represent synapomorphies) in beetles than between colour and shape of elytral spots, so we must base our exact weighting decisions on largely subjective evaluation, but anyway to ascribe them equal importance is evidently not adequate, and even the least distinctive weighting will be better than "none". The criteria to select weights may be various - individually variable characters, appearing frequently in various combinations in many groups of organisms, features functionally correlated to others (like fins and gills in water animals), those of simple genetical basis (e.g. albinism), are usually less informative than stable within species, unique for the taxon under study, functionally independent, genetically complex, &c. - but all resolve themselves to the assessment of the probability of homology (synapomorphy) vs. convergence (homoplasy).

Essentially the same holds true in case of overlapping characters: our decision is inevitably subjective, but as long as we know or suspect that they are genetically determined, i.e. may be phylogenetically informative, it is usually warranted to include them in the analysis. The exact form of the inclusion must, again, be subjectively selected according to "external evidence" (shape of distribution, degree of overlapping, possibility of statistical evaluation, &c.), constraints of the used program, and/or simply intuition; the most straightforward is subdivision of the entire range into convenient "segments" (4; or "logarithmically" 2-4 - 4-8 - 8-16 &c.) and ascribe to each analysed taxon the (calculated or estimated) average (or median, or modal, or minimum - as seems most appropriate in the given case) value. Anyway, the (unavoidably subjected...) assessment of adequacy is much more important than never really attainable (usually amounting to dogmatic schematism) "objectivity"!

Thanks for the paper on MICSEQ; I've downloaded it and begun to read it. I'll talk more about it later.

"How it is useful to give 'weight' to the particular character? Will this help more in resolving the tree?"

It can of course help in resolving the tree. The question is whether this resolution will be correct or spurious. Given that we don't know the development genetics of almost any morphological character, most characters could only be weighted based on wild guesses. The strategy most people use to avoid this obvious problem is to sample so many characters that the weight given to any single character becomes more or less irrelevant. This is what I recommend.

But nothing stops you from running your analysis twice, or five times, once with equal weights, the other times with any weighting scheme you like! Publish them all and compare the results (and the weighting schemes themselves) in your paper.

If you have few enough homoplastic characters in your matrix, you can try implied weighting, which downweights homoplastic characters, giving more impact to those characters that change states only once in the tree. But this condition is not always fulfilled. In my field, and in my own work, homoplasy-free characters are less common than those with moderate amounts of homoplasy.

True objectivity, whatever that is, is less important than reproducibility: your readers should be able to understand your coding and scoring decisions (why you chose certain characters and not others, why you divided them into states in a certain way and not others, why you scored particular states for particular taxa and not others, why you weighted certain state transitions in certain ways and not others...) even if they end up disagreeing with them.

"How to code quantitative character, if it is overlapping. e.g. number of leaflets. one species show 10-15 leaflets, another species show 12-18 leaflets. How to 'code' this character? Or just ignore such kind of characters for analysis?"

Don't ignore them! Don't throw information away.

I would divide this character into at least three states: 10–11 (state 0), 12–15 (state 1) and 16–18 (state 2) for example. Then, one species would have state 0 and 1, and the other would have states 1 and 2. It is not a problem to score more than one state of a character ("polymorphism") for any taxon.

(Obviously, in this case, you could declare each number of leaflets its own state. Then, one species would have states 0, 1, 2, 3, 4 and 5, and the other would have states 2, 3, 4, 5, 6, 7 and 8.)

Importantly, such a character would have to be ordered simply for logical consistency: if you use the assumption that it is easier to change from any condition to a very similar one than to a less similar one in order to divide the character into states in the first place, you should keep it when you weight the changes between the states of the character. This is the only case where objective weighting (within a character) is more or less possible, and should therefore be used.

12 XI 2018

David: “Given that we don’t know the development genetics of almost any morphological character, most characters could only be weighted based on wild guesses” – even this is better than "no" weighting: “guesses” of more or less experienced scientist are not based on soothsaying but on his/her knowledge (“external evidence”), and so almost “by definition” do contain additional information! But your basic assumption is also partly not true and partly irrelevant! We need not know the exact development genetics of any character, we need only to have some information as to its result in terms of the probability of change from “state” A to “state” B – and for this, as I mentioned in answer to Anup, we usually do have some clues: degree of variability within the analysed and other taxa, degree of correlation with other characters, probable (sometimes even perfectly evident) functional constraints, &c.!

“to sample so many characters that the weight given to any single character becomes more or less irrelevant” – on the one hand this mean the loss of much potentially useful information, on the other, so many characters are but rarely available.

“But nothing stops you from running your analysis twice, or five times, once with equal weights, the other times with any weighting scheme you like!” – this, again, is both usually impossible and only rarely desirable! The number of possible weighting schemes is (unless the numbers of taxa and characters are impracticably low) so enormous that it would be practically impossible to perform (to say nothing of publishing!) such a study, and at that we would end in completely uninterpretable multitude of formally legitimate trees without any reliable cue to their relative evaluation! In other words, we would continue to know “nothing” (to have many equally supported hypotheses is practically equal to having no hypothesis at all)!

“If you have few enough homoplastic characters in your matrix, you can try implied weighting, which downweights homoplastic characters, giving more impact to those characters that change states only once in the tree” – such “a posteriori weighting” is in fact a circular reasoning: we build a tree based on equally weighted transformations, and then give more weight to those supporting that particular tree!

“readers should be able to understand your coding and scoring decisions” – in rough outline it is usually either evident or done in description of the applied procedure; in more precise terms its is practically impossible and, in my opinion, not warranted (“average” reader will anyway not analyze such methodological details, while for those who wish to repeat the analysis according to their preferences important is rather whether the result is similar to mine, and which of them fits better the “external” evidence (biiogeographical pattern, ecological specificity of targeted taxa, &c.) – only after having clarified these points he/she would perhaps wish to look on which (presence/absence, definition, difference in weighting of) characters might cause the disparate results; anyway, exactly why I had chosen this or that character-sampling or -weighting is in all situations I am able to imagine utterly irrelevant (of course it becomes interesting if he/she wishes to discuss the question with me, but in such case it is especially evident that to burden the published paper with detailed explications!) is superfluous...

In case you might be interested, I’m “attaching” below one of my recent (2016) papers where you can see my current “custom” as to the presentation of phylogenetic reconstructions.

It seems that never. Morphology alone is in no way sufficient for phylogenetic analysis.

https://www.ingentaconnect.com/content/iapt/tax/2010/00000059/00000001/art00002

https://www.researchgate.net/publication/240509775_Molecular_phylogeny_of_Acacia_subgenus_Phyllodineae_Mimosoideae_Leguminosae_based_on_DNA_sequences_of_the_internal_transcribed_spacer_region

Ivan: "Morphology alone is in no way sufficient for phylogenetic analysis" - I, Ivan Mikhailovich Pchelin, have spoken. Howgh!

That's unfair; Ivan is not even an author of the paper he cited.

However, "clear [morphological] synapomorphies remain to be identified" doesn't mean that there aren't any; often it just means that people haven't looked hard enough. No wonder; this is time-consuming work that isn't going to be funded by agencies that want to promote "innovation".

(Roman, I owe you answers on several other topics. I'll get to them as soon as I can.)

David: I hope to remain in contact! Best wishes!

Dear Roman! Of course, the morphology can and MUST be used in the phylogenetic analysis. By the way, a competence in the morphology frequently means a competence in taxonomy. In contrast, confidente in molecular methods does not have any information abour the author's confidence in the taxonomic background, terms and ideology - usualy an activity of the molecular "taxonomists" is a mechanistic following some previously proposed formal methods. But the taxonomists are AUTHORS of the keybooks, not USERS (e.g. of methods). As you know Russian, I can recommend you to see my very unserious lection "Кибертаксономическая парадигма и возможность универсальной концепции вида как повод поговорить о системном кризисе естественных наук"

https://www.youtube.com/watch?v=a-p3cHaEXFE&t=2027s

All the best from Buffalo!

Alexey

Hello Alexey! Thank you very much for the comment - and for the "unserious lection", although I have a - serious... - problem with it: it seems to be only voice, and my computer does not reproduce voice...

Anyway, thanks and best wishes!

Roman

@Martin Kollmar: I agree with you about the source of information for to produce an evolutionary hipothesis, the phylogenetic tree. You can use any kind of data coming from the organism (morphology, DNA sequences, proteins, behaviour, etc.).

But you said "Every method and approach has its advantages and disadvantages, the developments over the years show that older methods (e.g. UPGMA, NJ, MP) have far more disadvantages than advantages compared with more modern methods. I would highly recommend to use Bayesian methods, which are the most advanced and reliable today." Being older the method doesn't mean being wrong. Bayesian phylogenetics, as maximum likelihood, needs a molecular model of evolution. Morphology can't be easily modeled, and the most common model being used when analizing morphology, the Mkv model, fails completely. See the following article (in press in Systematic Biology) for a demonstration:

Article Morphological Data Sets Fit a Common Mechanism Much More Poo...

Parsimony (using equal weigths and implied weigths) is the best option for morphology, and many more characters.

Cheers,

Camilo

"I would highly recommend to use Bayesian methods, which are the most advanced and reliable today."

Even apart from the issue of modeling just mentioned, their behavior with realistic morphological datasets, especially paleontological ones (i.e. with lots of missing data that is distributed neither randomly nor by rate of evolution), remains very poorly understood. Parsimony lacks this particular problem altogether; it is more sensitive to long-branch attraction, but that's generally a lesser problem in morphological datasets. Discussion in here: https://peerj.com/articles/5565

(This paper came out too early to cite the one in the previous comment, but does briefly mention that problem.)

This approach has a number of insurmountable disadvantages. First, the researcher in this case must be concerned not with the real relationship between organisms and their genealogy but with the similarity of traits or characters in terms of cladistics. Compliance with the condition of monophyly, which is a key factor for phylogenetics, becomes insurmountable. At the same time, in my opinion, the search for methods for the complex use of cladistic and semogenetic approaches is one of the most promising directions in the methodology of phylogenetic studies.

@ Martin Kollmar: I think networks are not today equally forward-looking for viruses, bacteria and, for example, vertebrates.

See also https://besjournals.onlinelibrary.wiley.com/doi/abs/10.1111/2041-210X.12760

Thank you Martin! Look, your evidence is compelling, but I am afraid that we all speak about a little different things and use terms in a slightly different way. Do you not think that the answer to the original question ("How I can say that the tree based only on morphological data is 'phylogenetic'?") will depend, inter alia, on the terminology and school of taxonomy (phenetics, cladistics etc)?

It depends most upon your characters being made of homologous structures.

If so, any method (distance, likelihood, parsimony, etc.) will produce a (more or less accurate/precise) phylogenetic hypothesis.

Hope it helps.