Chalmers Conferences, 9th European Conference on Mathematical and Theoretical Biology

Quantifying the effects of adaptation on genealogies
Taylor Austin Kessinger

Last modified: 2014-04-01


Population geneticists are often interested in inferring a population's history from observed patterns of diversity. Natural selection affects both: accordingly, it should distort a population's genealogy in measurable ways, and it should therefore be possible to infer selection based on a population's genealogy. Classical methods for inferring selection, such as Tajima's D and Fay and Wu's H, rely on noisy branch length estimates, depend strongly on demography, or become powerful only when many genealogies corresponding to different parts of the genome are available: on the other hand, classical tree "imbalance" metrics, such as the Sackin and Colless indices, are not sensitive to the types of distortions induced by selection.

We have developed three novel metrics for quantifying the effects of selection on a population's genealogy. Because they depend only on the topology of a particular tree, they are not subject to the pitfalls of classical methods. We develop statistical tests and demonstrate that selection can robustly be inferred based on one genealogy rather than a suite thereof.

In coalescent theoretical terms, a neutrally evolving population's genealogy is well described by the Kingman coalescent, whereas that of a rapidly adapting population is better described by the Bolthausen-Sznitman coalescent: populations between these two extremes can be thought of as arising from a Beta coalescent process, of which these two coalescents are specific instances. We regress against our statistics to determine which Beta coalescent process best describes a given population. We apply our method to the highly imbalanced genealogies characteristic of influenza: in general, this method should be appropriate for a range of asexual and facultatively sexual organisms.


Genealogies; adaptation; inferring selection