Metagenomic Binning

Groups contigs based on genomic signatures—intrinsic sequence features that are conserved within a genome but vary between species.

• Tetranucleotide frequency (TNF) : The most common signature, capturing the frequency of 4-mer words. Organisms have distinct TNF profiles driven by codon usage, restriction-modification systems, and replication mechanisms.
• GC content: Used as a coarse filter; contigs from the same genome typically share a narrow GC% range.
• Tools like MaxBin and MetaBAT use TNF and GC as input features for probabilistic models.
• Limitation: Fails to separate closely related strains that share similar compositional signatures.

Leverages the fact that contigs originating from the same genome co-vary in their relative abundance across multiple samples.

• Co-abundance profiles: If two contigs consistently increase or decrease in coverage together across dozens of samples, they likely belong to the same organism.
• CONCOCT uses Gaussian mixture models on combined composition and abundance features.
• Multi-sample requirement: This approach requires a time series or a set of related samples; it cannot function on a single metagenome.
• Strength: Can resolve strain-level differences when compositional signals are too similar.

Combines the outputs of multiple individual binning algorithms to produce a single, higher-confidence set of MAGs.

• DAS Tool: A consensus approach that calculates a single score per contig from multiple binner outputs, then optimizes bin assignments using a greedy algorithm.
• MetaWRAP: A modular pipeline that runs several binners, then uses a refinement module to consolidate bins by evaluating completeness and contamination.
• Ensemble logic: Reduces the systematic biases inherent in any single algorithm by requiring agreement from orthogonal methods.
• Result: Typically yields more MAGs with higher completeness and lower contamination than any single binner alone.

Uses neural networks to learn complex, non-linear representations of contig features without relying on hand-crafted genomic signatures.

• VAMB: A variational autoencoder that learns a latent representation from TNF and co-abundance, then clusters contigs in the learned embedding space.
• SemiBin: Employs a siamese neural network trained with contrastive learning, using reference genome data to teach the model which contigs should be binned together.
• Advantage: Deep models can capture subtle phylogenetic signals that linear methods miss, improving binning of low-abundance organisms and viral genomes.
• Hardware: Requires GPU acceleration for training on large metagenomic datasets.

Evaluates the completeness and contamination of reconstructed MAGs using universally conserved, single-copy marker genes.

• CheckM: The gold-standard tool that identifies lineage-specific marker gene sets and estimates completeness as the fraction of expected markers found, and contamination as the presence of duplicate markers.
• MIMAG standards: The Minimum Information about a Metagenome-Assembled Genome defines quality tiers—High-quality (≥90% complete, ≤5% contamination, with rRNA genes) and Medium-quality (≥50% complete, ≤10% contamination).
• BUSCO: An alternative approach using Benchmarking Universal Single-Copy Orthologs from the OrthoDB database.
• Critical insight: A MAG with 70% completeness and 10% contamination may be biologically misleading despite passing a threshold.

Uses read mapping coverage profiles to manually curate and refine automatically generated bins.

• Anvi'o: An interactive platform that visualizes contig coverage across samples as a hierarchical clustering dendrogram, allowing manual splitting or merging of bins.
• mmgenome: An R package for extracting, visualizing, and refining individual population genomes from metagenomes using coverage and essential gene content.
• GC vs. Coverage plots: A classic visualization where contigs from the same genome form distinct, tight clusters.
• Workflow: Automated binning is treated as a first pass; manual refinement using coverage data is essential for publication-quality MAGs.

A primary compositional signature used to bin contigs. Each microbial genome exhibits a characteristic frequency pattern of 4-base oligonucleotides due to species-specific codon usage, restriction-modification systems, and DNA structural preferences.

• Calculated as the frequency of all 256 possible 4-mers in a sliding window
• Emergent self-organizing maps (ESOMs) visualize TNF clustering
• Most effective for contigs >5 kbp; resolution degrades below 1 kbp
• Often combined with abundance profiles in differential coverage binning

A binning strategy that leverages abundance variation across multiple related samples. Contigs originating from the same genome will exhibit highly correlated read coverage patterns when mapped against independent metagenomes.

• Requires co-assembly of reads from multiple samples
• Coverage profiles are calculated by mapping reads from each sample back to the assembly
• Algorithms like CONCOCT and MetaBAT 2 integrate TNF and coverage using Gaussian mixture models
• Enables separation of closely related strains that share similar composition

An expectation-maximization algorithm that bins metagenomic contigs by fitting a probabilistic model of tetranucleotide frequencies and scaffold coverage levels.

• Assumes each genome's TNF and coverage follow a distinct multinomial and Poisson distribution
• Automatically estimates the number of bins using a modified Akaike Information Criterion (AIC)
• Incorporates single-copy marker genes as seeds to initialize the EM algorithm
• Effective for recovering genomes from low-abundance populations when coverage signal is weak