Gaussian process models for geographic controls in phylogenetic trees

Abstract

Background: Geographical confounding in phylogenetic inference models has long been an issue. Often models have great difficulty detecting whether congruences or similarities between languages in phylogenetic datasets stem from common genetic descent or geographical proximity effects such as language contact. Methods: In this study, we introduced a distance-based Gaussian process approach with latent phylogenetic distances that can detect potential geographic contact zones and subsequently account for geospatial biases in the resulting tree topologies. Results: We found that this approach is able to determine potential high-contact areas, making it possible to calculate the strength of this influence on both the tree-level (clade support) and the language-level (pairwise distances). Conclusions: Gaussian Process models are useful tools to address geographical confounding in phylogenetic analyses. The approach has advantages and disadvantages similar to comparable distance-based approaches, yet with this method, the level of confounding can be gauged and linguistic contact zones can be identified.