Employing phylogenetic tree shape statistics to resolve the underlying host population structure

Area of Science:

• Epidemiology
• Computational Biology
• Population Genetics

Background:

• Host population structure significantly influences infectious disease transmission.
• Phylogenetic trees offer insights into epidemic population structures.
• Identifying population structure guides the selection of appropriate phylogenetic methods.

Purpose of the Study:

• To develop and evaluate a classification procedure for distinguishing between structured and non-structured host populations using phylogenetic tree statistics.
• To assess the performance of machine learning classifiers in identifying population structure from simulated phylogenetic trees.

Main Methods:

• Simulated phylogenetic trees from structured and non-structured host populations.
• Computed eight tree statistics (e.g., number of cherries, Sackin index, Colless index, ladder length).
• Classified trees using Decision Tree (DT), K-nearest neighbor (KNN), and Support Vector Machine (SVM) algorithms, incorporating the basic reproductive number and performing sensitivity analyses.

Main Results:

• Machine learning classifiers achieved high accuracy (AUC > 0.9) in distinguishing between structured and non-structured populations.
• Support Vector Machine (SVM) models demonstrated greater robustness to variations in model parameters and tree size compared to DT and KNN.
• The classification procedure accurately identified a structured population in real-world data using an SVM-polynomial classifier.

Conclusions:

• The developed classification procedure effectively differentiates between structured and non-structured populations.
• SVM classifiers offer superior robustness for population structure analysis.
• The method shows high accuracy when applied to real-world epidemiological data, aiding in the understanding of disease spread.