Ancestral processes with selection (Q1370650)

It is shown how to construct the genealogy of a sample of genes for a large class of models with selection and mutation. Each gene corresponds to a single locus at which there is no recombination. The genealogy of the sample is embedded in a graph which was called as ancestral selection graph. This graph is the analogue of Kingman's coalescent process which arises in the case with no selection. The main goal of the paper is to analyze the ancestral selection graph and to compare it to Kingman's coalescent process. At the beginning the authors give an alternative formulation of the Moran model in terms of the biased voter model on a complete graph. Taking the diffusion limit and using duality, they construct an object which is analogous to the coalescent and which they call the ancestral selection graph. It is interesting to note that at first glance, the ancestral selection graph looks like the graph one obtains in the case of recombination without selection [\textit{R.C. Griffiths} and \textit{P. Marjoram}, IMA Vol. Math. Appl. 87, 257-270 (1997; Zbl 0893.92020)]; the way the graphs are interpreted, however, is very different in the two cases and there are basically no further similarities between the selection case and the neutral case with recombination. Further the authors analyze the ancestral selection graph, with and without mutation and give an algorithm for its simulation. Two cases of particular interest are addressed in the following corollary, namely the case when the selection coefficient (\(\sigma\)) is small, and the case when the sample size is large. It is shown how selection affects quantities obtained from the ancestral graph. To pick out the correct genealogical tree for a sample, a procedure for simulating the joint distribution of the types of a sample of size n is introduced and it is explained how to obtain the genealogy of the simulated sample. A procedure to compute the probability that two individuals in a sample are identical by descent is also discussed. This can be completed easily in the neutral case. But with the selection it is much more complicated and the authors are able to obtain the first two terms in its power series about \(\sigma=0\). The same method can also be used to obtain the power series about \(\sigma=0\) of the Laplace transform of the time to the most recent common ancestor for a sample of two individuals. In order to simulate backward along the sample path of the ancestral selection graph a scheme was applied using Markov chain Monte Carlo methods for approximating sampling probabilities via similar recursions for the neutral case, and this scheme was modified to the selective case. In the concluding section, the paper discusses another than the Moran model, and shows that they yield essentially the same ancestral selection graph. For models with selection the distribution of this process, the coalescent process, depends on the distribution of the frequencies of alleles in the ancestral generations. In contrast, the ancestral selection graph obtained in this paper does not require this knowledge. All one needs to know is the distribution of the frequencies of alleles at the time of the ultimate ancestor. It is not suprising that one needs to know the distribution of the frequencies of alleles throughout the ancestral history if one sample uses Kingman's coalescent to study the ancestral history of genes in the selective case, because Kingman's coalescent holds strictly only in the neutral case. Using the correct ancestral selection graph allows one to find the embedded genealogy and does not require knowledge of the frequency distribution at all times.