Bivariate Plot (3 points): Two variables plotted against one another
onan x,y coordinate graph. Correlation or regression analysis can be used
toobtain correlations. Correlations may be used to compare two variables,potentially
eliminating character dependency problems in a phylogeneticanalysis.
Procrustes analysis (3 points): This is a least squares fittingprocedure.
A configuration of landmarks may be used as a reference, while asecond
configuration is scaled, translated and rotated until the sum of thesquared
differences in the positions of homologous landmarks is as small aspossible.
The procrustes distance coefficient obtained from the analysis ofrelationships
between pairs of configurations can then be used as a measureof their difference.
A difference matrix (from comparisons of differentlandmark configurations
may then be assembled and analyzed using PCA orsome other multivariate
Thin Plate Spline (TPS)(3 points): TPS is a multivariate morphometricapproach that allows the morphometrician to use principal warps as featuresdescribing landmarks from a reference configuration. It's a kind of shapesummarization method and a good way to generate independent, potentiallyhomologous characters for use in phylogenetic analyses.
Configurations of landmarks may be broken down into linear (affine)
nonlinear (nonaffine) parts. Principal warps are eigenvectors (independent,
nonlinear functions) taken from the configuration of landmarks. The
nonlinear elements of the landmark configurations are TPS's of weighted
sums of principal warps. (see Rohlf, 1990 and various Bookstein
2) The pheneticist, cladist, and the evolutionary Biologist (6
On reconstruction of relationships:
P = overall similarity will resolve relationships
C = special similarity -- synapomorphy
EB = overall similarity in "adaptive specializations"
3) UPGMA tree's branch lengths will not agree with branch lengths computeddirectly form the distance matrix-- UPGMA (unweighted pair group methodusing arithmetic averages) starts with a table of distances between OTU's(the distances are measured in various ways - one example being numbers ofsubstitutions per sequence position). The two closest OTU's are found andjoined - the branch length (or branch depth) between them is taken to behalf the value of their distance difference. Once joined, the two taxa arethen merged into a cluster in a modified distance matrix and theirdistances to all other taxa are averaged. The original distance data arereplaced by averages (each taxon in the original data table contributesequally to the averages - hence, unweighted).
4) Agglomerative versus divisive clustering methods. As simply put aspossible, agglomerative clustering is a pairwise procedure which startswith a set of separate OTU's (or any number of separate "entities") andthen groups stepwise until a single set containing all OTU's obtains.Divisive methods start with all OTU's in a single set (star-like network)and subdivides into subsets.
The concepts are simple, but there may be some confusion as to what bestexemplifies these clustering techniques. UPGMA is a good example of theagglomerative method. However, NJ may not be a good example of divisive. InKim et al., 1993 you will find the NJ algorithm referred to as similar tothe agglomerative clustering methods described by Sneath and Sokal, 1973.This is because of the following: NJ begins with matrix of distances amongOTU's and an unresolved "star-tree." The algorithm pulls out the twoclosest pairs after averaging them together to create an HTU. Eachsubsequent HTU will represent the two previous closest pairs until only oneOTU remains and the star phylogeny is totally resolved. As such it is astep-wise/pairwise method.
Some good examples of divisive methods include dissimilarity analysis(MacNaughton-Smith et al., 1964) and association analysis (Williams andLambert, 1960, 1961) although these techniques are very "old school".
Take a look at Sokal and Sneath's Numerical Taxonomy (1973) - perhaps thebest reference you will find regarding this question.
5.) Good taxonomic character (20 points):
b) independent of other characters
c) discreet states (a system of at least two transformational homologs)
d) character varies more among than within OTU's
e) varies at a reasonable rate (not too fast, not too slow) (Albert
f) good potential for putative homology (i.e., complexity, many potentialcharacter states -- note difficulty in finding homologous sites in sequenceanalysis)
[+ Molecules vs. morphology for each]
6.) Compatibility analysis vs. Parsimony. (5 points) The largest cliqueof compatible characters will automatically remove all homoplasticcharacters and generate a single tree. One tree, no homoplasy - by the ruleof clique analysis, characters can change only once. Parsimony deals withall of the characters, but merely minimizes homoplasy.
A character is compatible if it needs no more than its minimum number ofevolutionary steps (k-1) when fitted in turn to all character state treesdrawn from each other character in the analysis. If there are x numbers ofcharacters (binary), there will be x number of binary character statetrees. Therefore, compatibility analysis may be using only a small fractionof the available characters and thus differ from the parsimony tree.
7.) Short definitions (10 points)
DNA hybridization: This 'whole genome' summary approach is intrinsicallya distance measurement. DNA is obtained and then sheared to get singlestranded DNA. That strand is radioactively labeled with radioactive iodineand hybridized with a "tester" DNA. if they are similar, they will anneal.Then, the hydrid DNA is denatured by heating. Temperature a% DNA break upare monitored. The temperature at which 50% of the DNA breaks up is ameasurement index of how similar the strands are.
This is phenetic technique. No individual characters. No assessments ofpositional homology. No understanding of character independence, etc.Step Matrix: Weighting character states based on extrinsic a priori data.For example, transitions are known to occur more frequently thantransversions, G&C are known to be more prevalent, 3rd positions changemore readily than 1st or 2nd, etc. these extrinsic pieces of evidence maybe plugged into the matrix as a state - weight scheme. (cf. Albert, et al.,1992)
Manhattan distance: "City block" distances between character states.Using manhattan distance over Euclidean ensures that one will pass throughreal data space when reconstructing ancestral/nodal characters (HTU's).
Retention Index: A measurement of character fit to a tree (established byFarris). Better than a CI (minimum possible length of tree/actual treelength), since CI does not remove autapomorphies (which have an automaticCI of 1.0) and is highly correlated with the number of taxa in a data set.The retention index = g-s/g-m; the minimum number of steps in a starphylogeny minus the total number of steps on parsimony tree divided by theminimum number of steps in a star phylogeny minus the minimum number ofsteps possible (number of character states -1; k-1). The retention index isthe fraction of apparent synapomorphy in the character that is retained assynapomorphy on the final tree.
Transversion: purine (A,G) <--> pyrimidine (C,T) changes
Character weighting: Increasing the "vote" of one character over another- done best on the basis of extrinsic evidence. Depends on a hypothesis ofgreater probability of change in some characters relative to others.
Epistemology: the HOW as opposed to the WHY (ontology). Empiricism.
Alternating sister group law. When using a series of outgroups - if thefirst and last outgroup have the same state, that state is the mostparsimonious assignment to the outgroup node; if they differ, the decisionis equivocal. (see Maddison et al., pg. 88)
Dollo parsimony: Gains happen once only and losses by reversal areallowed to explain the data (that, parallel or convergent gains of derivedstates cannot be invoked). Requires specification of polarity (to know whatstates are gained), and is generally considered unnecessarily strict andunrealistic.
Median state Rule: When creating a hypothetical taxonomic unit (HTU) fromits three joined taxa in a tree, the state assignments are based onmajority rule. If the three connected states for a particular character are110 then the state for the HTU will be 1. If all states differ (012, forexample) the median value or intermediate state will be assigned to the HTU(1).
8) Maximum likelihood vs. Max Parsimony:
Inductive vs. deductive. The latter is an optimality criterion which prefers minimal tree length. It isa deductive summary of the data matrix which (when taken as a model of aphylogenetic tree) assumes that apparent homology is more likely to be truehomology, than homoplasy. If characters are heritable, independent, andvarying at some informative rate (less than the rate of bifurcation - so asto avoid the "Felsenstein zone" (cf. question 10), for example), it followsthat one change on one branch is more likely than two on differentbranches. (Mishler, 1994)
Max likelihood is an inductive statistical procedure that maximizes theprobability of observing the data obtained (the sequences at the tips of aphylogeny), with respect to some explicit model of evolution. ML thusrequires additional assumptions about the data to maximize the probabilityof its occurrence given the tree. (refer to Chapter 11 of MolecularSystematics (Hillis et al., 1996) pg. 428 for a review of Olsen's lectureon ML! Also take a look at Siddall and Kluge. 1997. Probabilism andphylogenetic inference. Cladistics 4: 313-336 for a very interesting lookat this dichotomy in systematic approaches! Also take a look at JohnHuelsenbeck's web site (http://mw511.biol.berkeley.edu/)
Strict vs. majority consensus: The former will collapse all branches inconflict within a set of most parsimonious trees. The latter will keepthose branches which are consistent among 50% or more of the mostparsimonious trees. The argument for majority rule consensus that itconserves potentially important data (potentially true monophyleticgroups).
Taxic vs. Transformational homology: The former are synapomorphies -that is shared, derived features of the OTU's in question. Transformational homologies are those which are evident in ancestral-descendant lineages.Taxic homologies reveal monophyly while transformational homologies revealpolarity. If you shift your time horizon however, a transformational homology may be taxic or a taxic transformational.
Decay vs. Bootstrap: Decay indices report the degree to which parsimonycould be relaxed before particular monophyletic groups collapse. If a certain monophyletic group is present in all trees 5 steps longer than themost parsimonious tree, but not in all trees 6 steps longer, then it issaid to have a decay index of 6. Bootstrapping is a statistical procedurethat randomizes the data matrix and redraws trees based on the reconfigured matrices. If a monophyletic groups appears in 90% of the permutations, it is said to have a bootstrap of 90%.
Lundberg Rooting vs Outgroup Rooting: Lundberg rooting uses an inferredancestor only to root the ingroup network (at the place where the ANC would join), thus is more local. Outgroup rooting incorporates the outgroups asOTUs in the analysis along with the ingroup OTUs, thus is more global.
9) Quantitative characters. (10 points) Difficult to use because we needevidence of discrete transformational homology, corresponding to discreteevents that occurred along lineages in the past and can serve as markersfor that branch in the present (=taxonomic character). Monophyly is based on a discrete event (bifurcation) and needs a discrete marker, but incontinuous characters this is hard to see. Need to use ANOVA to see if variation between OTUs is more than that within OTUs.
10) Problems with inconsistency. (10 points) Tree "C" -- long branchattraction
will cause two lower branches to join, due to accumulatedhomoplasy. This
problem arises when two or more non-sister branches aremuch longer than
intervening ones. The problem is worse at higher overall rates of change
and with fewer character states.