Wright-Fisher model

cosmos 6th December 2017 at 1:26pm
Population genetics

There are many variants.

See Population genetics

Assumptions:

  • Non-overlapping generations
  • ...

Haploid Wright-Fisher model with selection

The definition can be found here:

Definition (Haploid Wright-Fisher model with selection): In a panmictic, haploid population of constant size NN, where individuals are of type aa and AA: if generation at time tt consists of kk individuals of type aa, and NkN-k of type AA, then, according to the Wright-Fisher model with selection, the generation at time t+1t+1 is formed by NN individuals, each of which has a probability to be of type aa given by:

P(type a)=k(1+s)k(1+s)+NkP(\text{type a}) = \frac{k(1+s)}{k(1+s)+N-k}

and is of type AA otherwise. The process is called sampling with replacement, because we are, in effect, replacing each individual of the previous population by a new one, which follows a given distribution of alleles (type). ss is called the selection coefficient, and 1+s1+s is the fitness of type aa. If, we give a fitness 1+s1+s' to type AA, then we use

P(type a)=k(1+s)k(1+s)+(Nk)(1+s)P(\text{type a}) = \frac{k(1+s)}{k(1+s)+(N-k)(1+s')}

And one can see how this would be generalized for more possible types in the model.

The way this probability comes about is:

  • The denominator is just to normalize the probability
  • In the numerator, kk is the number of individuals of type aa. The factors (1+s)(1+s) and (1+s)(1+s') determine the relative average number of offspring per individual. By this I mean that average number of offspring of an type a individualaverage number of offspring of an type a individual=1+s1+s\frac{\text{average number of offspring of an type a individual}}{\text{average number of offspring of an type a individual}} = \frac{1+s}{1+s'}. The average number of offspring of type aa, for instance is P(type a)NP(\text{type a}) N, as it is for a Bernoulli distribution (in this case, for the number of aa-type individuals), or a multinomial distribution, if more than two types are being considered.

If s=0s=0 for all types, selection doesn't play a role, and the model describes genetic drift only.

Haploid Wright-Fisher model with selection and mutation

Also described here.

Starting from the same setup as above (for the Haploid Wright-Fisher model with selection), the definition for the model with mutation is:

Definition (Haploid Wright-Fisher model with 'selection and mutation): If there are kk individuals of type aa among parents (and NkN-k individuals of type AA), and we have mutation rates μ1\mu_1 for aAa \rightarrow A, and μ2\mu_2 for AaA \rightarrow a, then, the probability of type aa (also called the proportion of potential offspring, in frequentist language, used often in biology) is:

ψk=k(1+s)(1μ1)k(1+s)+Nk+(Nk)μ2k(1+s)+Nk\psi_k = \frac{k(1+s)(1-\mu_1)}{k(1+s)+N-k} + \frac{(N-k)\mu_2}{k(1+s)+N-k}

As above, as each of the individuals in the next generation (offspring) have a type independently following this distribution. The number of type aa offspring follows a binomial distribution Bin(N,ψk)Bin(N, \psi_k)

Fixation

Diffusion approximation

See page 326 in here for instance


See this question

https://link.springer.com/chapter/10.1007/978-3-319-52045-2_2