Sampling bias and model choice in continuous phylogeography: Getting lost on a random walk

Area of Science:

• Population Genetics
• Computational Biology
• Epidemiology

Background:

• Phylogeographic inference reconstructs organismal spread using genetic and geographic data.
• Continuous phylogeography often employs Brownian Motion Phylogeography (BMP), modeling spread as Brownian motion in space and time.
• BMP is susceptible to sampling biases, particularly from under-sampled regions.

Purpose of the Study:

• To evaluate the impact of sampling bias on phylogeographic reconstructions.
• To compare the performance of Brownian Motion Phylogeography (BMP) and the spatial Λ-Fleming-Viot (ΛFV) process.
• To determine the optimal model for different epidemiological scenarios.

Main Methods:

• Investigated the effects of sampling bias on BMP reconstructions.
• Explored using sequence-free samples to mitigate BMP sampling bias.
• Analyzed the spatial Λ-Fleming-Viot (ΛFV) process as an alternative model.
• Compared model applicability based on their underlying assumptions and robustness to sampling bias.

Main Results:

• BMP reconstructions are significantly affected by sampling biases.
• Adding sequence-free samples can reduce bias but requires prior spatial distribution knowledge.
• The spatial ΛFV process demonstrates robustness to sampling biases.
• BMP and ΛFV models have distinct applications: ΛFV for endemic spread, BMP for recent colonization/outbreaks.

Conclusions:

• Sampling bias is a critical issue in phylogeographic inference using BMP.
• The spatial ΛFV process offers a more robust alternative, especially for endemic spread.
• Model selection in phylogeography should consider the specific spread dynamics and potential sampling biases.