User:R. Eric Collins/MBL/Popgen

Peter Beerli

Population Genetics

• assumption of independence among sequences violated in a populatino
• Wright-Fisher
  • individual drams parent from previous generation at random
  • wait on average 2N to coalesce
  • geometric distribution assumes discrete non-overlappy generations
  • real populations do not necessarily behave, but we assume that it can be extrapolated
  • because all samples contain information about the MRCA (most recent common ancestor), even samples with few individuals can most often recover the same TMRCA as a large sample
  • problem when too many samples are collected relative to generation size (> sqrt(4N)) because coalescent simplification assumes that not more than 1 coalescence happens per generation
  • large samples coalesce on averge in 4N generations

• mutations-scaled population size
  • hard to disentangle, large pop/small mu = small pop/large mu
  • confounding between migration rate and divergence rate

• recombination
  • if you know where these are you can/should break them up into separate loci
  • netrecodon to generate simulated recombination sequences
  • in order for recombination to make big differences you need to have VERY high rates of recombination
    • at least 1 in 50 per generation in a 2-population model
    • when migration is included it doesn't even seem to matter

• unless you have time-series data, don't bother with estimating population size changes through time (e.g. skyline/skyride)
• thanks 5x5 migratory model is even a large model
  • simplify model if possible to improve confidence/power
  • "test hypotheses... don't on fishing expeditions"

• to get good estimates, you need: 1) a lot of data 2) a good computer
• people with lots of data often don't run analyses long enough to guarantee convergence
• F_ST and coalescence are based on same/similar assumptions so really one is not better than the other for recent divergence
• shape of population size over time can really affect coalescence but need to know how and how it affects parameter estimation
  • e.g. bottlenecks, recoveries, expansions, contractions
• when effective population size ~ generations since divergence it can get dicey to separate divergence from migration

• no existing coalescent program take selection into account

• felsenstein 2005: after ~10 individuals, should add another locus rather than more individuals

Species Tree Estimation

• with long times between speciation, the gene tree matches the species tree with increasing probability
• two ways to coalesce to ((A,B),(C,D)), one way each to coalesce (((C,D),B),A) and (((C,D),A),B)
  • so symmetric trees can be overrepresented
• concatenated gene sequences are not the way to add information, can lead to statistically inconsistent results
  • but with long branch lengths and lots of genes you get enough power that it's ok
  • Bootstrap procedure can be positively misled in this situation

• STEM: when only source of variability in single-gene histories is due to thecoalescence process

• species definition? a group of individuals that fit a model of random branching

• questions:
  • if order generations, can follow min and max to find ancestor of all existing species
  • migration as horizontal gene transfer?

"the reason to do a bayesian analysis is not to get a tree but to get the posterior distributions"

HGT versus huge ancestral population size + long coalescent times "bacteria are special"

LAMARC

• if there were exponential growth in a population, estimating the mutation rate assuming a constant population size will UNDERESTIMATE the instantaneous mutation rate.
• given the instantaneous mutation rate and a population growth model you can estimate the past mutation rate