Deep clustering of bacterial tree images

Area of Science:

• Genomic epidemiology
• Microbial genomics
• Evolutionary biology

Background:

• Whole-genome sequencing (WGS) is increasingly used in pathogen surveillance programs.
• Phylogenetic trees visualize pathogen ancestry, relatedness, and diversification.
• The 'comet' tree shape of methicillin-resistant Staphylococcus aureus (MRSA) aids interpretation of drug resistance evolution.

Purpose of the Study:

• To identify and categorize novel phylogenetic tree shapes in bacterial WGS datasets.
• To understand the evolutionary and epidemiological implications of different tree shapes.
• To expand the interpretive framework beyond the known MRSA 'comet' shape.

Main Methods:

• Extraction of phylogenetic trees from large bacterial WGS datasets.
• Visualization of trees as images for pattern recognition.
• Clustering of tree images to identify common shapes and groups.

Main Results:

• Identification of nine major groups of phylogenetic tree shapes.
• Discovery of shapes including 'comets', star-like phylogenies, and 'barbell' phylogenies.
• Association of these shapes with potential evolutionary and epidemiological narratives.

Conclusions:

• Bacterial WGS data exhibit diverse phylogenetic tree shapes beyond the MRSA 'comet'.
• These shapes offer insights into microbial evolution and population structures.
• Further research can leverage these patterns for enhanced pathogen surveillance and understanding.