Gene-wide identification of episodic selection

Abstract

We present BUSTED, a new approach to identifying gene-wide evidence of episodic positive selection, where the non-synonymous substitution rate is transiently greater than the synonymous rate. BUSTED can be used either on an entire phylogeny (without requiring an a priori hypothesis regarding which branches are under positive selection) or on a pre-specified subset of foreground lineages (if a suitable a priori hypothesis is available). Selection is modeled as varying stochastically over branches and sites, and we propose a computationally inexpensive evidence metric for identifying sites subject to episodic positive selection on any foreground branches. We compare BUSTED with existing models on simulated and empirical data. An implementation is available on www.datamonkey.org/busted, with a widget allowing the interactive specification of foreground branches.

Keywords: branch-site model; episodic selection; evolutionary model; random effects model.

Fig. 1.

Depiction of our online widget used to interactively specify foreground branches, where such a priori information is available. This example is from a subset of the HIV-1 RT data set (Murrell, de Oliveira, et al. 2012), where terminal branches leading to samples taken after antiretroviral therapy are selected as foreground. Results for this analysis can be seen in table 2. The widget facilitates the annotation of large trees like this one (only a small subtree is shown for legibility), for example, by labeling all branches using a text pattern (e.g., here all branches of interest start with a “T”), and allowing automatic labeling of internal branches (e.g., using parsimony labeling).

Fig. 2.

Statistical performance of BUSTED and fitmodel. (A) Power of BUSTED as a function of ω₃, for various nominal significance levels. The weight assigned to ω₃ by the model was 0.1. See text for other simulation parameters. (B) Type 1 error rate as a function of the nominal significance level (null data), showing that BUSTED is conservative and fitmodel is anticonservative. (C) Power of BUSTED and fitmodel as a function of simulated selective strength (ω₃), using test significance levels set to achieve 0.05 Type I error rate on null simulations (fitmodel was anticonservative).

Fig. 3.

Correlates of signal for episodic selection in the selectome data sets. Each panel depicts the fraction of all alignments reported by BUSTED as positively selected (at $P\le 0.05$), as a function of (A) the length of the alignment (codons), censored at 2000 due to sparse sampling afterwards, (B) the number of sequences, (C) the total tree length (expected number of substitutions per codon site), (D) the maximum-likelihood estimate of the ω₃ parameter, used as a proxy for the “strength” of selection. Plot points were chosen through an adaptive binning scheme, with each point representing at least 100 data sets. Lowess smoothing polynomials (smoothing span 0.25) are shown in solid light gray.