DeepMicrobes

DeepMicrobes applies deep learning to classify metagenomic data for taxonomic identification at species and genus levels and to estimate relative abundances without relying on curated taxonomic trees.

Key Features:

• Deep learning-based taxonomic classification: Uses deep learning models to assign taxonomic labels to metagenomic reads and assemblies.
• Taxonomy-independent classification: Classifies metagenomic data without depending on traditional curated taxonomic trees or the integration of genome assemblies into reference databases.
• Species and genus identification: Demonstrates superior performance in species- and genus-level identification compared to state-of-the-art tools.
• Abundance estimation: Maintains comparable accuracy in estimating microbial relative abundances.
• Training data: Trained on genome assemblies reconstructed from gut microbiomes.
• Novel signature discovery: Enables discovery and characterization of potential novel microbial signatures linked to conditions such as inflammatory bowel diseases.

Scientific Applications:

• Taxonomic profiling: Assign species- and genus-level taxonomic labels to metagenomic reads and assemblies.
• Relative abundance estimation: Estimate microbial relative abundances in metagenomic samples.
• Novel species characterization: Identify and characterize potential novel microbial signatures from genome assemblies, including those reconstructed from gut microbiomes.
• Microbiome research in health and environment: Investigate roles of uncultured or unrepresented species in human health contexts (e.g., inflammatory bowel diseases) and environmental habitats.

Methodology:

DeepMicrobes employs deep learning models trained on genome assemblies reconstructed from gut microbiomes to classify metagenomic data without relying on traditional curated taxonomic trees.