Bayesian inference of species hybrids using multilocus dominant genetic markers

Abstract

Neutral genetic markers are useful for identifying species hybrids in natural populations, especially when used in conjunction with statistical methods like the one implemented in the software NEWHYBRIDS. Here, a short description of the extension of NEWHYBRIDS to dominant markers is given. Subsequently, an extensive series of simulations of amplified fragment length polymorphism (AFLP) data is performed to evaluate the prospects for hybrid identification with (possibly non-diagnostic) dominant markers. Distinguishing between F1's and F2's is shown to be difficult, possibly requiring upwards of 100 AFLP markers to be done accurately. Discriminating between pure-bred and non-pure (hybrid) individuals, however, is shown to be much easier, requiring perhaps as few as 10 dominant markers, even from relatively weakly diverged species.

Figure 1

Directed graphs describing the NewHybrids model. Black-shaded nodes represent observed data, grey-shaded nodes represent parameters of prior distributions and unshaded nodes represent unobserved variables. The boxes are ‘plates’ which denote multiple, conditionally independent replicates (indexed by the subscript ℓ over loci and i over individuals in the sample) of the enclosed nodes. (a) The model for codominant data. Note that the observed data are the allelic types Y_i,ℓ,1 and Y_i,ℓ,2 of the two gene copies at each locus in each individual (hence those nodes are shaded black). (b) The model for dominant data: each dominant locus is modelled as a diallelic locus with a recessive allele and a dominant one. These allele types are no longer observed, so they are latent variables. At each dominant locus, the observed datum is the presence or absence of a band, $Y_{i,\ell }^{\text{dom}}$.

Figure 2

Posterior mean estimates (y-axis) versus true values (x-axis) of the frequency of the recessive r allele from either species gene pool (A or B) across all simulated datasets (both LD and HD conditions) with H=100 and L=100.There are 20 000 points in total, each one corresponding to an allele frequency at a single locus in species A or B.

Figure 3

Mean posterior probability of hybrid category membership. (a–f) The mean over all simulations with different simulation scenarios denoted by the plot symbols as follows: circles, LD and H=100; triangles, LD and H=1000; pluses, HD and H=100; crosses, HD and H=1000. The x-axis of (a–f) shows L, the number of markers used. (a–f) Correspond to different hybrid categories as indicated on each plot. Note that the y-axis scale differs on each plot: (a) A, (b) F₁, (c) ${\text{BC}}_1^{\text{A}}$, (d) B, (e) F₂, (f) ${\text{BC}}_1^{\text{B}}$.

Figure 4

ROC curves based on P(Pure) (see text) computed from simulation output for the LD scenario with H=100 and L equal to 10 (long-dashed line), 25 (short dashed line) and 50 (solid line). A separate ROC curve has been plotted for each true hybrid category, i.e. (a) shows the rate at which pure A individuals are allocated to the pure category as a function of the rate at which individuals from the non-pure categories are incorrectly assigned to the pure category; (b) shows the rate that F₁'s are allocated to the non-pure category as a function of the rate at which pure A and B individuals are incorrectly assigned to the non-pure category, etc. (a) A, (b) F_1, (c) ${\text{BC}}_1^{\text{A}}$, (d) B, (e) F₂, (f) ${\text{BC}}_1^{\text{B}}$.