Discrete trait analysis (DTA) methods are widely used to infer patterns of movement between discrete geographic locations from genetic sequence data as they are simple and fast to implement. With the increasing interest in phylogeography for pathogen genomic epidemiology, accounting for and understanding the limitations of the methods used is becoming increasingly important. Applying DTA to location makes several strong simplifying assumptions, including the assumption of equal sampling probability across locations. This may lead to biased estimates of migration rates and ancestral locations where sampling rates vary. Here we investigate these biases using simulated migration and transmission data. We propose a simple correction to DTA for unequal sampling probability across locations when sampling rates are known or can be estimated. This correction to the migration rate matrix can be applied only to phylogenetic branches leading to sampled tips or to the whole phylogeny. We assess the performance of this correction using simulated data for a range of endemic transmission scenarios with varied sampling and migration rates. In a simple, two-location, case the correction performs significantly better than a standard DTA method in three of the four sampling rate distributions investigated. Applying the correction across only the tips of the phylogeny and over the full depth of the tree gives similar results, with the correction only at the tips performing marginally better in this simple two-location simulation. As migration is a key driver of infectious disease spread, improving estimates of migration with simple phylogenetic methods is a valuable development in infectious disease phylogeography.

A simple correction to adjust for sampling biases in phylogeographic discrete trait analysis