Nominal and ordinal predictors in logistic regression for the identification of adaptive loci in animal and plant populations

When searching for loci possibly under selection in the genome, an alternative to population genetics theoretical models is to establish allele distribution models for each locus to directly correlate allelic frequencies and environmental variables like precipitation, temperature, or sun radiation for example. Such an approach implementing multiple logistic regression models in parallel was implemented within a computing programme named MATSAM. Recently, this application was improved in order to support qualitative environmental predictors as well as to permit the identification of associations between genomic variation and individual phenotypes, allowing the detection of loci involved in the genetic architecture of polymorphic characters. Here we present the corresponding methodological developments and compare the results produced by software implementing population genetics theoretical models (DFDIST, BAYESCAN) and allele distribution models (MATSAM) in an empirical context to detect signatures of genomic divergence associated with speciation in Lake Victoria cichlid fishes.