Abstract for the talk at 15.10.2013 (15:15 h)VW-Seminar
Omri Tal (MPI MIS, Leipzig)
Quantifying Information on 'Population Structure' from Genetic Data
See the homepage of the speaker.
I propose a framework to derive measures of 'informativeness' for population structure from multi-locus genetic data. Specifically, given a collection of genotypes sampled from known multiple populations I would like to quantify the potential for correct classification of genotypes of unknown origin, or alternatively, provide a measure of data 'clusteredness'. Motivated by Shannon's axiomatic approach in deriving a unique information measure for communication, I first identify a set of intuitively justifiable criteria that any such quantitative information measure should satisfy, also suggesting that the notion of communication noise in this framework is analogous to sampling noise. I will show that standard information-theoretic measures such as mutual information or relative entropy cannot satisfactorily account for this sense of information, necessitating methods from statistical-learning. I will also review very recent empirical work of biologists to assess the 'population signal' from genetic samples.