mge-cluster

mge-cluster implements a reference-free classification framework for clustering and typing mobile genetic elements using unitig presence/absence profiles, openTSNE dimensionality reduction, and HDBSCAN clustering.

Key Features:

• Reference-free classification: The framework classifies plasmids without requiring reference databases.
• Unitig compression: Input sequences are compressed using a unitig representation in a de Bruijn graph.
• Scalable processing: The approach handles thousands of input sequences with faster runtime than existing algorithms and moderate memory usage.
• Dimensionality reduction: openTSNE is used for dimensionality reduction of unitig presence/absence profiles.
• Clustering: HDBSCAN is used for clustering mobile genetic elements.
• Novel type recognition: The approach simultaneously recognizes novel types and classifies plasmids into previously identified groups.
• Consistent labelling: The framework enables consistent labelling of plasmids across past, present, and future sequence collections.

Scientific Applications:

• Plasmid population analysis: Population-wide plasmid data set analysis from Escherichia coli.
• Resistance gene prevalence studies: Studying the prevalence of the colistin resistance gene mcr-1.1 within plasmid populations.
• Resistance plasmid transmission tracking: Describing instances of resistance plasmid transmission within hospital environments.

Methodology:

Sequences are compressed using a unitig representation in a de Bruijn graph, followed by openTSNE dimensionality reduction and HDBSCAN clustering on unitig presence/absence profiles.