Skip to content
Tech Interview Prep
Technical interview guide

Phylogenetics & Evolutionary Analysis

Reconstructing evolutionary relationships from sequence data: tree-building methods, substitution models, and molecular clocks.

Read
50 min
Practice MCQs
25
Interview Q&A
25
Edition
v3
Editorial status
Reviewed

Scope: MAFFT 7; current IQ-TREE, RAxML-NG, MrBayes and BEAST documentation; ModelFinder, ultrafast bootstrap, multispecies coalescent and recombination source literature reviewed 2026-09-04.

Curated: · Written: · Reviewed:

25 questions

MCQ-10A team reconstructs a phylogeny from 40 morphological characters scored across 12 taxa. No substitution model exists for these characters, so they run maximum parsimony. Which criterion does that method optimize when it ranks candidate trees?
MCQ-11A colleague runs 1,000 nonparametric bootstrap replicates on a 2,000-site alignment and reports a clade with 95% bootstrap support. They write in the manuscript that "there is a 95% probability this clade is real." A reviewer objects. Which statement best describes what the 95% actually measures in this analysis?
MCQ-1A 60-sequence alignment of a conserved kinase domain is built with MAFFT. Column 112 contains 58 hydrophobic residues (L/I/V) and 2 gaps; column 113 contains 30 gaps and 30 residues of mixed chemistry. The team runs IQ-TREE with a substitution model that treats gaps as missing data. Which statement best describes what the alignment asserts at column 112?
MCQ-15You fit a GTR+Γ model to a 2,000-bp alignment of 12 primate species. The branch leading to humans has length 0.07. The human-chimp raw p-distance is 0.01, and no fossil calibration or molecular clock was used. What does the 0.07 on that branch represent?
MCQ-14An unrooted maximum-parsimony tree for four ingroup taxa has topology ((A,B),(C,D)). At a binary site, the states are A=0, B=0, C=1, D=1. You root the tree with an outgroup O that attaches to the branch leading to A and has state 0. What does this rooting imply about the ancestral state at the root and the number of changes on the tree?
MCQ-19A gene family has four copies in each of three sequenced species. The gene tree is rooted with an outgroup and shows two deep clades, each containing one copy from every species; within each clade the species' sequences group by species, not by clade. A colleague concludes the duplications happened after the three species split. What does the tree actually support?
MCQ-25A comparative study has 60 mammal species on a time-calibrated tree rooted at 60 Myr. Lineages were sampled at different times, so root-to-tip path lengths run from 10 to 60 Myr and the tree is not ultrametric. A trait is simulated under an additive Brownian motion process: each Myr along a lineage adds an independent, unbiased variance increment of sigma^2 per Myr, with no directional trend and no pull toward an optimum. For a tip whose root-to-tip path length is t, what does the expected variance of its trait value, measured from the root value, do as t grows?
MCQ-5You fit two substitution models to the same 5,000-site, four-taxon alignment (observed base frequencies A = 0.35, T = 0.35, C = 0.15, G = 0.15). Both recover the same topology, but JC69 estimates a total tree length of 0.42 substitutions per site while GTR estimates 0.28. Which explanation accounts for the gap?
MCQ-8Two labs build phylogenies for the same genus from different data. Lab A aligns 2,200 sites across 18 taxa and fits GTR+Γ; its best tree has lnL = −9,120. Lab B aligns 1,500 sites across 12 taxa and fits HKY+Γ; its best tree has lnL = −8,950. A reviewer concludes that Lab B's tree is better supported because its likelihood is higher. Which assessment of that conclusion is correct?
MCQ-18Three species A, B and C are each sequenced at 1,000 single-copy orthologous loci. The expected species tree is ((A,B),C), whose internal branch is estimated at 0.5 coalescent units. Locus trees from the 1,000 alignments split 596 for ((A,B),C), 202 for ((A,C),B) and 202 for ((B,C),A). What do these counts indicate for this data set, and what should be done next?
MCQ-20Loci are sampled from three diploid taxa whose species tree is ((A,B),C), one lineage per taxon. The internal branch running from the A-B split to the split with C lasted 100,000 generations, and the effective size of the population on that branch was 1,000,000. The loci are unlinked and neutral, the terminal branches are long, and no gene flow between lineages is detected. What fraction of loci is expected to have a gene tree that conflicts with ((A,B),C)?
MCQ-2A team is building a species tree for 15 genomes from 480 gene families, each family kept because it has exactly one copy in every genome. In family G, the deepest split of the rooted gene tree separates genomes 4 and 9 from the other 13 at full support. Copies within each of those two groups differ at 5-7% of third-codon positions, the two groups differ at 24%, and the deepest split among the 15 species is about 12%. The outgroup's two copies of family G, A and B, fall one on each side of that split. Every other family puts genome 4 with genome 11. What should the team conclude about family G?
MCQ-22You sequence 120 single-copy core genes and the recA locus for 40 isolates of genus X. The 120-gene concatenation splits them into two clades that match their source environment. The recA tree instead places 6 clade-A isolates inside genus Y, sampled from the same environment. Those 6 recA alleles are 99.2% identical to the genus Y allele and 87% identical to the other clade-A recA alleles, and each lies immediately downstream of an integrase gene and a tRNA-Arg gene. What does this pattern indicate, and which tree should be the species tree?
MCQ-21A team aligns 12,000 bp from 20 bacterial genomes and infers a maximum-likelihood tree. Strain X sits with clade A (bootstrap 71). They then split the alignment into four 3,000-bp blocks and infer a tree from each. In blocks 1 and 2, X groups with clade A (bootstrap 98 and 96); in blocks 3 and 4, X groups with clade B (bootstrap 99 and 97). Per-block branch lengths for X are 0.03–0.05 substitutions/site; the concatenated tree gives X 0.14. What is the most defensible reading and next step?
MCQ-6An alignment of 2,000 simulated sites was generated under GTR+Γ with four discrete rate categories (α = 0.5) and no invariant sites. Fitting GTR+I+Γ to that alignment raises the log-likelihood by 22 over GTR+Γ, with estimated p_inv = 0.31 and α = 0.78. The ML topology is unchanged. What does the improvement most likely reflect?
MCQ-9A 20-taxon, 4,000-site alignment is analysed by maximum likelihood. One heuristic search from a neighbor-joining start, using NNI rearrangements until no swap improves the likelihood, converges at lnL = −18,204. A rerun with 100 random starting trees and the same NNI stopping rule finds a best of lnL = −18,190, reached by 7 of the 100 runs; the other 93 end at −18,204. Which conclusion is best supported?
MCQ-7A 9,000-column concatenated alignment of six protein-coding genes is fitted under two partition schemes. Scheme A splits by gene and codon position (18 partitions); scheme B splits by gene only (6 partitions). Every partition gets GTR+Γ: 5 relative exchangeability rates, 3 base frequencies and 1 gamma shape, 9 free parameters, and branch lengths are shared and identical in both schemes. Scheme A reaches lnL = −24,318; scheme B reaches lnL = −24,502. Partitions were fixed from the gene annotation before any tree was estimated. The team's protocol selects on BIC with n = 9,000. Which scheme does it select?
MCQ-12A team runs Bayesian MCMC on a 5-taxon alignment under GTR+G and reports a clade with posterior probability 0.97. A reviewer reruns the same alignment under JC69 and gets 0.62 for that clade. Both runs used the same four chains, the same seed, and converged (split R-hat 1.01, ESS > 500). The clade is present in every sampled tree in the GTR+G run. What does the drop to 0.62 most directly show about the 0.97?
MCQ-13A team fits a two-component mixture model with four chains, 2,000,000 iterations each, discarding the first 500,000 as burn-in. The four chains' posterior means for the mixing weight are 0.31, 0.30, 0.62 and 0.61; the split-R-hat is 1.9 and the effective sample size is 38. The team reports convergence because each chain ran for two million iterations. Which conclusion is correct?
MCQ-17A 40 Ma fossil is the oldest known member of clade A, so the divergence at the base of A cannot be younger than 40 Ma. An analyst calibrates that node in BEAST 2 with a normal prior, mean 40 Ma, sd 5 Ma, and the analysis returns a node age of 34 Ma. Which change fixes the calibration?
MCQ-16A team dates a 12-taxon primate alignment using a strict molecular clock in BEAST 2 and gets a 95% HPD for the root age of 18–34 Myr, far wider than the 23–25 Myr implied by the fossil record. A colleague reruns the same alignment and priors with an uncorrelated relaxed clock (lognormal), and the root HPD narrows to 22–26 Myr. The alignment has 4 partitions, and a stepping-stone marginal-likelihood estimate gives a log marginal likelihood of -8420 for the strict clock and -8395 for the relaxed clock. What does the relaxed clock change about the model that best explains the narrower interval?
MCQ-3Parsimony on a 22-taxon, 6,000-site nuclear matrix recovers fast-evolving ingroup taxon A (ML terminal branch 0.29 substitutions/site) as sister to outgroup taxon D (0.26 substitutions/site) at 96% bootstrap; GTR+G ML instead places A beside A1 and A2 (0.01 substitutions/site each) inside genus C and D beside its own congeners. Every other terminal branch is 0.005-0.02 substitutions/site. You add four specimens sequenced for the same 6,000 sites; a three-gene analysis places all four in the crown of genus C, on branches that are not on the lineage leading to A. You re-run the same parsimony search on the 26-taxon matrix. Where does A come out?
MCQ-4A 12-taxon, 5,000-column protein alignment is run twice. On the untrimmed alignment, maximum likelihood groups taxa A and B with 68 percent bootstrap support over 1,000 replicates. An automated similarity-based column filter then keeps 3,100 columns, and of the alignment's 1,200 parsimony-informative sites only 180 remain; the same search on the trimmed alignment puts (A,B) at 96 percent and moves taxon G to a different position. The trimmed tree is the one written up. Which conclusion about the (A,B) clade do the two runs support?
MCQ-23A 42-taxon nucleotide matrix is analysed under GTR+G. Two taxa, X and Y, each have estimated branch lengths about three times the median, and they form a clade with 100% bootstrap support. X and Y are 91% identical to each other but at most 78% identical to any other taxon, and both are AT-rich relative to the rest of the matrix. When 12 slowly evolving taxa that break the X and Y branches are added, the clade disappears and X and Y attach to different parts of the tree. What is the most defensible reading of the original 100% support?
MCQ-24A 512-taxon maximum-likelihood tree is written as Newick with an integer from 0 to 100 in the label slot of every internal node, as in (A:0.004112,B:0.011904)87:0.002331, and every branch length to six decimals. The producer's report lists 214 clades at or above 70 percent support. At a second lab, a figure viewer and a filtering script both read that slot as clade names, so the script's support column comes out empty on the same file. Which change removes the disagreement for both labs?