Metagenomic Assembly and Binning: A Practical Workflow for Recovering Genomes from Complex Microbial Communities

Metagenomic assembly and binning convert raw sequencing reads from mixed microbial communities into draft genomes that can be analyzed without cultivation. This workflow matters for researchers who need to recover metagenome-assembled genomes (MAGs) from environmental, clinical, or agricultural samples. The practical goal is to select tools and parameters that match your data type and community complexity, then execute a reproducible pipeline from quality control through genome binning. This article provides a decision framework for that process, with emphasis on the tradeoffs between short-read, long-read, and hybrid approaches.

Scope and Reader Context

This workflow applies to researchers, analysts, and life-science professionals who have already generated or plan to generate metagenomic sequencing data. You should be comfortable with command-line environments and have access to a Linux server or high-performance computing cluster. The guidance covers the core steps of read preprocessing, assembly, contig binning, and post-assembly quality assessment. It does not cover experimental design for sample collection, library preparation, or sequencing platform selection in depth, though these choices influence downstream analysis outcomes.

The central challenge in metagenomic assembly is that you are reconstructing genomes from a mixture of organisms with varying abundance, sequence similarity, and genome complexity. Assembly combines short input reads into longer contiguous fragments called contigs. Binning then clusters these contigs into individual genome bins that represent putative organisms. Both processes have algorithmic limitations that manifest differently depending on whether you process each sample separately or pool multiple samples together. A fully automated workflow called Metaphor demonstrates this flexibility by offering multiple assembly and binning strategies combined with a bin refinement step to achieve high-quality genome bins. The workflow also generates reports to evaluate its own performance across different synthetic datasets.

Core Principles of Metagenomic Assembly

Read Quality Filtering as the First Decision Point

Raw sequencing reads contain adapter sequences, low-quality bases, and potential contamination from the sequencing platform. Read preprocessing is the first step in any metagenomic assembly workflow because assembly quality depends directly on input read quality. A routinely used metagenomic assembly workflow begins with read quality filtering before any assembly attempt. This step removes low-quality reads and trims adapters, which reduces the computational burden of assembly and improves contig accuracy.

The filtering stringency should match your downstream goals. Aggressive filtering removes more data but may discard short reads from low-abundance organisms. Lenient filtering preserves more data but introduces more errors into the assembly graph. For complex communities with many low-abundance taxa, you may need to balance sensitivity against error rate. The decision should be recorded in your analysis log so that results can be compared across samples processed with different parameters.

Assembly Algorithms and Their Assumptions

Assembly algorithms reconstruct genomes by building graphs that represent overlaps between reads. Short-read assemblers such as MEGAHIT and metaSPAdes use de Bruijn graph approaches that are computationally efficient but sensitive to sequencing errors and repetitive regions. Long-read assemblers such as Flye use overlap-layout-consensus approaches that handle longer reads but require higher sequencing depth and have different error profiles.

The choice of assembler depends on your data type and community complexity. Short-read assemblers work well for Illumina data and are the default choice for most metagenomic projects. Long-read assemblers require Oxford Nanopore or PacBio data and can resolve repetitive regions that short reads cannot span. Hybrid assembly combines both data types to leverage the accuracy of short reads and the contiguity of long reads. The MUFFIN workflow demonstrates this hybrid approach by using short and long reads together to produce high-quality bins and their annotations, with the workflow orchestrated in Nextflow for reproducibility.

Community Complexity and Sequencing Depth

Community complexity refers to the number of distinct organisms in a sample and their relative abundances. A simple community with a few dominant taxa assembles more easily than a complex community with hundreds of low-abundance species. Sequencing depth determines how many reads cover each genome and directly affects assembly completeness. Low-depth samples produce fragmented assemblies because assemblers cannot resolve the full genome structure.

Benchmarking studies show that sequencing depth and taxonomic complexity strongly impact binning performance. Simulated benchmarking datasets often exhibit substantially lower complexity than real human gut and environmental metagenomes, which means results from simulated data may not transfer directly to your samples. You should assess your community complexity before selecting assembly parameters and consider whether your sequencing depth is sufficient for the organisms you want to recover.

At a Glance: Tool Selection Decision Table

The following table compares commonly used assemblers and binners with criteria for tool selection based on data type and community complexity.

Multi-sample binning consistently outperforms single-sample binning across short-read, long-read, and hybrid data according to a benchmark of 13 binning tools. The same benchmark found that multi-sample binning is more effective at identifying potential antibiotic resistance gene hosts and near-complete strains containing biosynthetic gene clusters. If you have multiple samples from the same study, you should consider co-assembly or multi-sample binning strategies instead of processing each sample independently.

Practical Workflow from QC to Genome Binning

Step 1: Read Quality Control and Preprocessing

Start with raw sequencing reads in FASTQ format. Run quality assessment tools to evaluate per-base quality scores, GC content, adapter contamination, and duplication rates. Remove adapters and trim low-quality bases using parameters appropriate for your sequencing platform. For paired-end short reads, verify that read pairs remain synchronized after trimming.

Record the number of reads before and after filtering, the total base count, and the quality metrics. These numbers become part of your analysis provenance and allow you to compare results across samples. The DOE JGI Metagenome Workflow performs read filtering as the first module in its processing pipeline, and this workflow scales to thousands of metagenome samples per year with varying community complexity and sequencing depth.

Step 2: Assembly Strategy Selection

Choose your assembly strategy based on data availability and research questions. For short-read data only, select between MEGAHIT and metaSPAdes based on your compute resources and contiguity needs. For long-read data, use Flye or a comparable long-read assembler. For hybrid data, use a workflow that integrates both read types, such as MUFFIN, which was tested on biogas reactor samples and produces functional pathway predictions alongside genome bins.

Consider whether to assemble each sample separately or co-assemble multiple samples. Co-assembly can improve assembly of shared organisms across samples but may produce chimeric contigs for strain-variant organisms. The Metaphor workflow offers flexible approaches for assembly and binning of input data, allowing you to choose between per-sample and pooled strategies based on your study design.

Step 3: Contig Processing and Coverage Calculation

After assembly, filter contigs by length to remove short fragments that are unlikely to represent meaningful genomic content. A minimum contig length threshold of 1000 base pairs is common, though the optimal threshold depends on your community and assembly quality. Calculate coverage for each contig by mapping the original reads back to the assembly. Coverage information is essential for binning because most binning algorithms use coverage profiles alongside sequence composition.

For multi-sample datasets, calculate coverage for each sample separately and combine these profiles into a matrix. This differential coverage approach improves binning because organisms with similar abundance patterns across samples are more likely to be grouped together. A comparison of single-coverage and multi-coverage binning on the same set of samples demonstrated that multi-coverage binning produces better results and identifies contaminant contigs and chimeric bins that single-coverage approaches miss.

Step 4: Genome Binning

Select a binning tool based on your data type and sample structure. MetaBAT2 is a robust default for short-read data. MaxBin2 works well for single-sample datasets. CONCOCT is appropriate for multi-sample differential coverage approaches. For long-read or hybrid data, use binners that accept these input types or convert your assembly to a format compatible with your chosen tool.

Run the binner and examine the output bin count and bin quality metrics. Most binners produce a set of bins with associated completeness and contamination estimates. These estimates are typically based on the presence of single-copy marker genes. A bin with high completeness and low contamination represents a near-complete genome. The MAGqual pipeline provides an accessible way to evaluate metagenome quality and generate metadata on a large scale, built in Snakemake for readability and scalability.

Step 5: Bin Refinement and Reassembly

Initial binning results often contain bins that are incomplete, contaminated, or chimeric. Bin refinement tools such as BinSPreader improve binning quality by using assembly graph information to reassign contigs. The BinSPreader protocol describes steps for a typical metagenome assembly and binning workflow, then details binning refinement variants, output, and possible caveats. This refinement step optimizes the process of reconstructing more complete genomes from metagenomic data.

For bins that are incomplete but have low contamination, consider reassembly. The MetaflowX workflow includes a dedicated reassembly module that improved MAG quality in benchmarking tests, increasing completeness by 5.6% and reducing contamination by 53% on average. Reassembly extracts reads that map to a target bin and assembles them independently, which can resolve genome structure that was lost in the whole-community assembly.

Step 6: Quality Assessment and Reporting

Assess final bin quality using standardized criteria. The MIMAG standards provide a framework for categorizing genome quality, though these standards have not been widely adopted according to the MAGqual pipeline developers. MAGqual evaluates metagenome quality against MIMAG standards and generates metadata on a large scale.

Report the number of bins recovered, their completeness and contamination estimates, and the taxonomic assignments. Include assembly statistics such as N50, total assembled bases, and number of contigs. These metrics allow other researchers to evaluate the quality of your genome recovery and compare it to other studies. The ATLAS workflow provides abundance estimates at genome resolution for each sample in a dataset, which is useful for comparative analyses across samples.

Options and Tradeoffs in Assembly and Binning

Per-Sample versus Multi-Sample Processing

The decision to process samples individually or together has major consequences for genome recovery. Per-sample processing is simpler and requires fewer compute resources, but it may miss organisms that are shared across samples or present at low abundance in individual samples. Multi-sample processing leverages information from multiple samples to improve assembly and binning, but it requires more compute and storage.

Benchmarking evidence supports multi-sample binning as the superior approach. A comparison of single-coverage and multi-coverage binning on the same set of samples demonstrated that multi-coverage binning produces better results and identifies contaminant contigs and chimeric bins that other approaches miss. The study authors concluded that multi-coverage binning should always be performed over single-coverage binning when analyzing multi-sample datasets.

The optimal number of samples for multi-sample binning is approximately 20. Using too few or too many samples can reduce the benefits of multi-sample binning according to a comprehensive benchmark of binning tools. If you have fewer than 5 samples, the added complexity of multi-sample binning may not be justified. If you have more than 50 samples, consider grouping samples by environment type or experimental condition.

Short-Read, Long-Read, and Hybrid Approaches

Short-read sequencing remains the most common approach for metagenomics due to its low cost and high accuracy. However, metagenome-assembled genomes from short-read data are often fragmented, particularly for complex communities. Long-read sequencing technologies promise to improve genome reconstruction by spanning repetitive regions, but the integration of two different sequencing modalities makes downstream analyses complex.

Hybrid approaches combine short and long reads to achieve both accuracy and contiguity. The MUFFIN workflow uses short and long reads to produce high-quality bins and their annotations, with taxonomic classification and KEGG pathway predictions for each bin. This workflow was tested on twenty biogas reactor samples and demonstrated the capacity to process and output relevant files for microbial community analysis.

The choice between these approaches depends on your research questions and budget. If you need complete genomes for novel organisms, long-read or hybrid approaches are worth the additional cost. If you are surveying community composition or recovering partial genomes for abundant taxa, short-read data may be sufficient.

Assembly-Based versus Alignment-Based Analysis

Metagenomic analysis can be divided into assembly-based and alignment-based approaches. Assembly-based approaches reconstruct genomes from the reads and then analyze the assembled contigs. Alignment-based approaches map reads directly to reference databases and analyze the mapping results. Assembly-based approaches provide higher taxonomic and functional resolution but require heavy computational resources. Alignment-based approaches are fast and scalable, but their performance depends on the comprehensiveness of the reference database.

For fungal metatranscriptomics, standardized comparisons show assembly-based pipelines deliver higher taxonomic and functional resolution but require heavy computational resources, while alignment-based pipelines are fast and scalable with fungal-specific marker-based tools performing better. The same tradeoff applies to metagenomics generally. If your community contains many novel organisms without close relatives in reference databases, assembly-based approaches are necessary.

Observations and Measurements

Recording Assembly and Binning Metrics

Systematic record keeping is essential for reproducible metagenomic analysis. Record the following metrics for each sample and analysis run:

• Raw read count and base pairs before filtering
• Filtered read count and base pairs after quality control
• Assembly time and peak memory usage
• Number of contigs and total assembled bases
• Contig N50 and L50 values
• Number of bins recovered
• Completeness and contamination estimates for each bin
• Taxonomic assignments for each bin

These records allow you to compare results across samples, identify problematic samples, and report your methods accurately. The DOE JGI Metagenome Workflow uses 20 publicly available sediment metagenomes to illustrate the computing requirements for different steps and highlight typical results of data processing. You can use similar reference datasets to benchmark your own workflow performance.

Benchmarking Your Workflow

Before processing your full dataset, run your workflow on a small subset or a publicly available benchmark dataset. This practice identifies parameter issues and resource bottlenecks before you commit significant compute time. Benchmarking studies provide reference points for expected performance. A benchmark of 13 binning tools across short-read, long-read, and hybrid data under co-assembly, single-sample, and multi-sample binning found that multi-sample binning exhibits optimal performance across all data types.

For short-read data specifically, a comprehensive benchmark of binning tools using simulated and real metagenomic datasets found that neural network-based tools consistently outperformed other tools in genome recovery from both real samples and simulated samples with realistic taxonomic complexity, though at higher computational cost. If you have access to sufficient compute resources, consider testing neural network-based binners in addition to traditional approaches.

Interpreting Quality Metrics

Completeness and contamination estimates are the primary quality metrics for genome bins. Completeness estimates the fraction of single-copy marker genes present in the bin. Contamination estimates the fraction of marker genes that appear in multiple copies, indicating that the bin may contain sequences from multiple organisms. A bin with completeness above 90% and contamination below 5% is generally considered a high-quality draft genome.

These estimates have limitations. They rely on the assumption that marker genes are single-copy in the target organism, which may not hold for all taxa. They also depend on the completeness of the marker gene database used for estimation. You should interpret these metrics as estimates instead of exact measurements and consider additional evidence such as taxonomic consistency and coverage uniformity when evaluating bin quality.

Common Failure Patterns

Fragmented Assemblies

Fragmented assemblies produce many short contigs that cannot be binned effectively. This pattern occurs when sequencing depth is insufficient, community complexity is high, or the assembler parameters are not optimized for the data. Symptoms include low N50 values, a high number of contigs below the length threshold, and few bins with high completeness.

To address fragmented assemblies, increase sequencing depth, adjust assembler parameters, or switch to a different assembler. For short-read data, metaSPAdes may produce longer contigs than MEGAHIT at the cost of higher memory usage. For long-read data, ensure sufficient depth and consider error correction before assembly.

Chimeric Bins

Chimeric bins contain contigs from multiple organisms, producing high contamination estimates. This pattern occurs when the binner groups contigs from closely related organisms or when coverage profiles are not sufficiently distinct. Symptoms include contamination estimates above 10%, inconsistent taxonomic assignments within a bin, and variable GC content across contigs in the same bin.

To address chimeric bins, use bin refinement tools that leverage assembly graph information, such as BinSPreader. Multi-coverage binning can also identify contaminant contigs and chimeric bins that single-coverage approaches miss. If chimerism persists, consider using a different binner or adjusting binning parameters.

Low Genome Recovery

Low genome recovery means that few high-quality bins are produced from the dataset. This pattern occurs when the community is highly complex, sequencing depth is insufficient for low-abundance organisms, or the binning approach is not well matched to the data. Symptoms include few bins above the completeness threshold, many bins with high contamination, and poor recovery of known community members.

To improve genome recovery, consider multi-sample binning, which outperforms single-sample binning across data types. Integrating and refining genome bins from the top three binning tools recovered more than 30% more high-quality genomes than previous methods in one benchmark study. This approach requires more compute but can substantially improve results.

Parameter Sensitivity

Many assembly and binning tools are sensitive to parameter choices, and default parameters may not be optimal for your data. This pattern manifests as widely varying results when parameters are changed slightly. Symptoms include different bin counts and quality metrics across parameter combinations, and difficulty reproducing results across runs.

To address parameter sensitivity, document all parameters used in your analysis, test a small parameter grid on a subset of your data, and select parameters based on objective quality metrics instead of intuition. Workflow tools such as ATLAS and MetaflowX provide modular frameworks that help maintain provenance and reproducibility.

Limitations and Interpretation Boundaries

Strain-Level Resolution

Metagenomic assembly and binning typically recover genomes at the species level, not the strain level. Closely related strains within a species may be assembled into a single consensus genome or separated into multiple bins with shared contigs. This limitation affects downstream analyses such as functional annotation and comparative genomics.

Representation learning and graph-based binning methods have demonstrated high strain-level resolution and can reduce contamination in complex microbial communities. However, these methods are computationally intensive and may not be practical for all datasets. If strain-level resolution is critical for your research question, consider using long-read sequencing or targeted enrichment approaches.

Low-Abundance Organisms

Organisms present at very low abundance in the community are difficult to recover through assembly and binning. Their reads are a small fraction of the total dataset, and their genomes may be fragmented or absent from the assembly. This limitation is inherent to the approach and cannot be fully overcome by computational methods.

If you need to recover genomes from low-abundance organisms, consider increasing sequencing depth, using enrichment techniques, or applying targeted assembly approaches. The recovery of low-abundance taxa remains an active area of research, with artificial intelligence-based methods showing promise for improving quality control, error correction, assembly, binning, refinement, and annotation procedures.

Reference Database Dependence

Taxonomic classification and functional annotation of genome bins depend on reference databases. If your organisms are novel or distantly related to sequenced representatives, classification may be uncertain and functional annotations may be incomplete. This limitation is particularly relevant for environmental samples from understudied habitats.

Reference-based methods that compute beta-diversity distances between metagenomes are highly dependent on the quality and completeness of the reference database, and their application on less studied microbiota can be challenging. De novo comparative metagenomic methods that rely only on sequence composition can complement reference-based approaches for such datasets.

Quality and Reproducibility Controls

Workflow Management Systems

Workflow management systems such as Snakemake and Nextflow provide reproducibility through automated pipeline execution, dependency management, and provenance tracking. The ATLAS workflow is implemented in Snakemake and provides a user-friendly, modular, and customizable framework for metagenome data processing. The MUFFIN workflow uses Nextflow to achieve high reproducibility and fast, straightforward use.

Using a workflow management system ensures that your analysis is reproducible, which is essential for scientific validity and for compliance with data sharing policies. The NIH Genomic Data Sharing Policy outlines expectations for data sharing and reproducibility in genomic research. The FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable.

Version Control and Documentation

Record the versions of all software tools used in your analysis, along with their parameters and reference databases. This information should be included in your methods section and made available with your data. Version control systems such as Git can track changes to your analysis scripts and configuration files.

The complexity of computational workflows required to process metagenomic data includes dozens of bioinformatics software tools, each with their own set of customizable parameters that affect the final output. Documenting these parameters is essential for reproducibility and for interpreting differences between studies.

Data Management and Sharing

Metagenomic data should be deposited in public repositories such as NCBI, which provides data resources for sequence data and associated metadata. The EMBL-EBI Training program offers resources for learning about data submission and analysis. Data sharing policies may apply depending on your funding source and research context.

The NIH Genomic Data Sharing Policy describes expectations for sharing genomic data generated with NIH funding. The FAIR Guiding Principles provide a framework for making data findable, accessible, interoperable, and reusable. Following these principles improves the impact and reproducibility of your research.

Safety and Regulatory Context

Data Privacy and Security

Metagenomic data from human-associated samples may contain identifiable information and are subject to privacy protections. De-identify samples before sequencing and follow institutional review board requirements for human subjects research. Store raw sequence data and clinical metadata securely and control access according to your data sharing plan.

The NIH Genomic Data Sharing Policy outlines expectations for protecting research participant privacy while enabling data sharing for research purposes. If your samples come from clinical settings, consult your institutional privacy officer before depositing data in public repositories.

Computational Resource Management

Metagenomic assembly and binning require substantial computational resources. Large datasets can require hundreds of gigabytes of memory and days of compute time. Plan your resource allocation before starting your analysis and monitor resource usage during execution.

The DOE JGI Metagenome Workflow scales to run on thousands of metagenome samples per year, which can vary by the complexity of microbial communities and sequencing depth. The workflow modules for read filtering and metagenome assembly are available as a workflow description language file, and the annotation and binning modules are provided as a service to the user community.

Professional Escalation Criteria

Seek expert assistance when you encounter the following situations:

• Assembly fails repeatedly with different tools and parameters
• Binning produces no high-quality bins from a dataset that should contain recoverable genomes
• Quality metrics are inconsistent across replicates or runs
• You need to recover genomes from a community with known high complexity and low sequencing depth
• You are unsure whether your sequencing depth is sufficient for your research questions

Bioinformatics support teams, computational biology consultants, and experienced metagenomics researchers can provide guidance on troubleshooting and workflow optimization. The EMBL-EBI Training program and NCBI Data Resources offer educational materials and support for researchers at various skill levels.

Frequently Asked Questions

What is the difference between assembly and binning in metagenomics?

Assembly combines short input reads into longer contiguous fragments called contigs. Binning clusters these contigs into individual genome bins that represent putative organisms. Assembly reconstructs the sequence content of the community, while binning assigns that content to specific organisms. Both processes are at the core of genome-resolved metagenomics workflows, and their limitations pose different challenges depending on the selected strategy.

How do I choose between MEGAHIT and metaSPAdes for short-read assembly?

Choose MEGAHIT when you have large datasets, limited compute resources, or highly complex communities where memory efficiency is critical. Choose metaSPAdes when you need longer contigs and have sufficient memory available. metaSPAdes generally produces longer contigs than MEGAHIT but requires more memory and compute time. Test both tools on a subset of your data and compare assembly statistics before committing to a full run.

When should I use long-read or hybrid assembly instead of short-read assembly?

Use long-read or hybrid assembly when you need to resolve repetitive regions, recover complete or near-complete genomes from complex communities, or analyze organisms with large genomes or complex genomic structures. Long-read sequencing technologies promise to improve genome reconstruction, but the integration of two different sequencing modalities makes downstream analyses complex. Hybrid workflows such as MUFFIN use short and long reads together to produce high-quality bins and their annotations.

How many samples should I include in multi-sample binning?

Approximately 20 samples is the optimal number for multi-sample binning. Using too few or too many samples can reduce the benefits of multi-sample binning. Multi-sample binning outperforms single-sample binning across short-read, long-read, and hybrid data, and it is more effective at identifying potential antibiotic resistance gene hosts and near-complete strains containing biosynthetic gene clusters.

What completeness and contamination thresholds should I use for high-quality bins?

A bin with completeness above 90% and contamination below 5% is generally considered a high-quality draft genome. The MIMAG standards provide a framework for categorizing genome quality, though these standards have not been widely adopted. The MAGqual pipeline evaluates metagenome quality against MIMAG standards and generates metadata on a large scale.

How do I fix chimeric bins that contain contigs from multiple organisms?

Use bin refinement tools such as BinSPreader that leverage assembly graph information to reassign contigs. Multi-coverage binning can also identify contaminant contigs and chimeric bins that single-coverage approaches miss. If chimerism persists, consider using a different binner or adjusting binning parameters. Integrating and refining genome bins from multiple binning tools can recover more high-quality genomes than using a single tool.

What computational resources do I need for metagenomic assembly and binning?

Resource requirements depend on dataset size, community complexity, and the tools you select. Large datasets can require hundreds of gigabytes of memory and days of compute time. The DOE JGI Metagenome Workflow scales to run on thousands of metagenome samples per year, with computing requirements that vary by community complexity and sequencing depth. Start with a small subset of your data to estimate resource needs before processing the full dataset.

How do I ensure my metagenomic analysis is reproducible?

Use a workflow management system such as Snakemake or Nextflow to automate pipeline execution and track dependencies. Record the versions of all software tools, their parameters, and reference databases. Deposit your data in public repositories such as NCBI and follow the FAIR Guiding Principles for data management. The ATLAS workflow is implemented in Snakemake and provides a user-friendly, modular framework for metagenome data processing.

Related Bioinformatics Guides

• Long-Read Genome Assembly and Polishing Strategies
• Long Read Metagenomic Assembly: Structural Analysis and Computational Methodologies in Bioinformatics
• Transcriptome Assembly Without a Reference Genome
• Metagenomic Binning Strategies for the Animal Gut Microbiome
• Long-Read Sequencing Technologies: PacBio and Oxford Nanopore

References and Further Reading

• EMBL-EBI Training. European Bioinformatics Institute.
• NCBI Data Resources. National Center for Biotechnology Information.
• Genomic Data Sharing Policy. National Institutes of Health.
• The FAIR Guiding Principles. Scientific Data.
• Metaphor-A workflow for streamlined assembly and binning of metagenomes.. GigaScience, 2022.
• Metagenomics workflow for hybrid assembly, differential coverage binning, metatranscriptomics and pathway analysis (MUFFIN).. PLoS computational biology, 2021.
• Metagenomic Assembly: Reconstructing Genomes from Metagenomes.. Methods in molecular biology (Clifton, N.J.), 2021.
• Protocol for refining metagenomic binning with BinSPreader.. STAR protocols, 2023.
• ATLAS: a Snakemake workflow for assembly, annotation, and genomic binning of metagenome sequence data.. BMC bioinformatics, 2020.
• MetaflowX: a scalable and resource-efficient workflow for multi-strategy metagenomic analysis.. Nucleic acids research, 2025.
• A bioinformatician's guide to metagenomics.. Microbiology and molecular biology reviews : MMBR, 2008.
• DOE JGI Metagenome Workflow.. mSystems, 2021.
• Artificial intelligence in metagenome-assembled genome reconstruction: Tools, pipelines, and future directions.. 2026.
• Benchmarking alignment strategies for Hi-C reads in metagenomic Hi-C data.. 2026.
• Interpreting fungal ecological contributions through taxonomic and functional profiling of metatranscriptomics.. 2026.
• MAGqual: a stand-alone pipeline to assess the quality of metagenome-assembled genomes. bioRxiv, 2023.
• Benchmarking metagenomic binning tools on real datasets across sequencing platforms and binning modes. Nature Communications, 2025.
• Comprehensive benchmarking of metagenomic binning tools reveals key factors for improved genome recovery. Nature Communications, 2026.
• A comparison of single-coverage and multi-coverage metagenomic binning reveals extensive hidden contamination. Nature Methods, 2023.
• Comparison of k-mer-based de novo comparative metagenomic tools and approaches. Microbiome Research Reports, 2023.

This article is educational and does not replace validated analysis plans, institutional policy, clinical interpretation, or specialist review.