SemiBin

SemiBin applies semi-supervised deep learning to perform metagenomic binning, grouping contigs into metagenome-assembled genomes (MAGs) by integrating reference-genome information while retaining de novo binning capability.

Key Features:

• Semi-supervised neural network: Employs neural-network models to incorporate reference-genome information into the binning process.
• Reference integration with de novo capability: Integrates information from reference genomes while enabling recovery of genomes not present in reference databases.
• MAG reconstruction: Groups sequences predicted to originate from the same genome to construct metagenome-assembled genomes (MAGs).
• Improved bin quality: Demonstrates superior performance relative to state-of-the-art binning methods on simulated and real microbiome datasets.
• Cross-environment validation: Validated on human gut, dog gut, and marine microbiomes, yielding more high-quality bins and increased taxonomic diversity.
• Enhanced taxonomic recovery: Produces an increased number of distinct genera and species in assembled bins.

Scientific Applications:

• MAG reconstruction from metagenomes: Recovery of metagenome-assembled genomes from metagenomic assemblies.
• Microbial diversity profiling: Improved detection and recovery of diverse taxa, increasing identified genera and species.
• Comparative microbiome analysis: Application to human gut, dog gut, and marine microbiomes for cross-environment genome recovery and comparison.

Methodology:

Applies a semi-supervised neural-network model that integrates reference-genome information while maintaining de novo binning capability; evaluated on simulated and real microbiome datasets from human gut, dog gut, and marine environments.