Scimm

Scimm performs unsupervised clustering of metagenomic reads from environmental DNA sequencing using interpolated Markov models to group sequences by species origin.

Key Features:

• Unsupervised learning: Operates without training data and clusters reads based on intrinsic sequence patterns rather than reference genomes.
• Interpolated Markov models: Uses interpolated Markov models to capture sequence dependencies at multiple scales for improved discrimination of origin.
• Species-level read clustering: Groups reads that originate from the same species to support downstream taxonomic binning.
• Database independence: Does not rely on pre-existing sequenced genomes from public databases for clustering.
• PHY SCIMM hybrid approach: Combines SCIMM with supervised techniques such as Phymm to enhance clustering accuracy when evolutionarily close training genomes are available.
• Improved accuracy: Demonstrates higher clustering accuracy compared to earlier methods in reported evaluations.

Scientific Applications:

• Metagenomic binning: Assigns environmental sequencing reads to species-level clusters for reconstruction of community composition.
• Novel microbe discovery: Facilitates analysis of samples containing predominantly novel microbial taxa by not depending on reference genomes.
• Hybrid supervised-enhanced clustering: Uses PHY SCIMM to improve clustering accuracy in samples with representatives from well-characterized genera.

Methodology:

Unsupervised clustering using interpolated Markov models to analyze sequence patterns and cluster reads by similarity; PHY SCIMM integrates SCIMM with supervised Phymm classification.