Metagenome Co-Assembly: Strategies for Multi-Sample Data

Metagenome co-assembly is the process of combining sequencing reads from multiple samples into a single assembly run to recover shared community members and low-abundance organisms that individual assemblies may miss. This approach is appropriate when samples come from the same environment type, when the research question targets community-level gene content, or when computational resources are limited. Co-assembly trades per-sample strain resolution for broader community coverage, and the choice between co-assembly and individual assembly depends on sample relatedness, study objectives, and available computing infrastructure.

Understanding Co-Assembly in Metagenomic Workflows

Metagenomic assembly reconstructs microbial genomes from complex mixtures of DNA sequencing reads. When researchers have multiple samples from a related study, they face a fundamental decision: assemble each sample separately or combine all reads into one co-assembly. The EMBL-EBI Training materials describe assembly as a core step in metagenomic analysis pipelines, where quality-controlled reads are converted into contiguous sequences for downstream gene prediction and taxonomic classification.

The NCBI Data Resources provide access to the Sequence Read Archive and other repositories where metagenomic datasets are deposited. These public resources enable researchers to compare assembly strategies across studies and to access benchmark datasets for method development. The scale of publicly available metagenomic data, with over 600,000 metagenomes in repositories such as the Sequence Read Archive, highlights the need for reproducible assembly workflows that can handle large multi-sample projects 13.

Co-assembly differs from individual assembly in several important ways. In individual assembly, each sample is processed independently, producing sample-specific contigs that preserve the strain composition of that particular community. In co-assembly, reads from all samples are pooled and assembled together, which can recover genes that are too low in abundance to assemble from any single sample 6. The tradeoff is that combining samples increases the risk of mixing data from closely related strains, which can complicate the assembly graph and reduce contiguity 6.

Core Principles of Multi-Sample Assembly

When Co-Assembly Provides Clear Benefits

Co-assembly is most effective when samples share substantial community membership. Longitudinal studies of the same host or environment, replicate samples from the same site, and time series experiments all produce samples that are likely to share dominant taxa. In these cases, co-assembly increases sequencing depth for shared organisms, allowing the assembler to resolve genomes that would be fragmented in individual assemblies.

A study of the Baltic Sea using 124 samples compared three assembly strategies: individual assembly followed by gene clustering, co-assembly of all samples, and a mix approach combining both 6. The mix-assembly approach produced a more extensive nonredundant gene set than either individual assembly or co-assembly alone, with more genes predicted to be complete and functionally annotated 6. This finding suggests that a hybrid strategy can capture the benefits of both approaches.

When Individual Assembly Is Preferable

Individual assembly preserves the strain-level resolution of each sample. When samples come from very different environments, when strain-level differences are the focus of the study, or when samples have very different community compositions, individual assembly followed by gene clustering may be preferable 6. The risk of chimeric assemblies increases when reads from closely related but distinct strains are combined, and this risk is amplified in co-assembly.

For studies focused on structural variants or strain dynamics over time, co-assembly graphs can be used to detect genomic alterations across samples. A method called rhea constructs a single co-assembly graph from all samples in a series and calculates the log fold change in graph coverage between successive samples to identify structural variants that are increasing or decreasing in abundance 9. This approach demonstrates that co-assembly can support analyses beyond simple genome recovery.

Computational Resource Considerations

Traditional co-assembly of large datasets can require substantial memory and computing time. A sequential co-assembly method was developed to address this limitation by reducing the assembly of duplicate reads through successive application of single-node computing tools for read assembly and mapping 5. Using a simulated mouse microbiome dataset, this approach shortened assembly time, used less memory than traditional co-assembly, and produced significantly fewer assembly errors 5.

The sequential co-assembly method was applied to a 2.3-terabyte dataset from gnotobiotic mice that was too large for traditional co-assembly, demonstrating significant reductions in assembly time and memory requirements 5. This approach is particularly useful in resource-constrained settings, including in low- and middle-income countries 5.

At a Glance: Co-Assembly Decision Table

Practical Workflow for Co-Assembly

Step 1: Quality Control and Read Preprocessing

Before co-assembly, all samples must undergo quality control to remove adapter sequences, low-quality bases, and contaminant reads. The TOFU-MAaPO pipeline provides a portable, automated workflow for raw data pre-processing, assembly of metagenome-assembled genomes, and taxonomic and functional annotation 13. This pipeline can analyze metagenome files locally or directly from the Sequence Read Archive using accession or study IDs 13.

Quality control decisions affect downstream assembly quality. Reads with low average quality scores, high duplication rates, or contamination from host DNA should be filtered before assembly. The FAIR Guiding Principles emphasize that data should be Findable, Accessible, Interoperable, and Reusable, which applies to both raw reads and assembled contigs 4. Documenting quality control parameters ensures that assemblies can be reproduced and compared across studies.

Step 2: Selecting Assembly Tools

Two commonly used metagenomic assemblers are MEGAHIT and metaSPAdes. A benchmarking study of metagenome-assembled genome pipelines found that metaSPAdes consistently outperformed MEGAHIT in terms of recall or completeness 15. However, MEGAHIT is often faster and uses less memory, making it suitable for large datasets or resource-constrained environments.

The choice of assembler depends on the specific goals of the study. For gene catalogue construction, the assembler must produce contigs that capture the full coding potential of the community. For metagenome-assembled genome recovery, the assembler must produce long, contiguous sequences that support genome binning. The Methods in Molecular Biology chapter on metagenomic assembly and gene prediction outlines core assembly strategies, including per-sample versus co-assembly and short-read versus hybrid approaches, and highlights key parameters and metrics for evaluating assembly quality 8.

Step 3: Running the Co-Assembly

For co-assembly, all quality-controlled reads from the samples to be combined are provided as input to the assembler. The specific commands depend on the assembler and the computing environment. For MEGAHIT, reads from multiple samples can be provided as a list of files. For metaSPAdes, the --meta flag enables metagenomic mode, and multiple read libraries can be specified.

The nf-UnO pipeline implements metagenomic co-assembly in Nextflow for novel pathogen detection from metagenomic shotgun sequencing of epidemiologically related foodborne outbreak specimens 11. This pipeline uses MIDAS2, metagenomic co-assembly, multiple binning programs, and read mapping to metagenomically assembled genomes to detect potential etiological agents found in common across outbreak specimens 11.

Step 4: Assessing Assembly Quality

After co-assembly, the quality of the resulting contigs must be assessed. Key metrics include N50, which measures the contiguity of the assembly, the total number of contigs, the largest contig length, and the number of bases assembled. These metrics provide a summary of assembly quality but do not capture all aspects of assembly correctness.

The MAG-E framework provides a generalizable approach for end-to-end evaluation of entire metagenome-assembled genome pipelines, from assembly through binning to quality control and filtering 15. This framework relies on simulations that match an ecosystem of interest and provide ground truth for accurate evaluation 15.

Step 5: Binning and Genome Recovery

Co-assembled contigs are typically binned into metagenome-assembled genomes using tools that group contigs by sequence composition and coverage across samples. Multi-sample binning is most effective with about 20 samples, as using too few or too many samples can reduce its benefits 14. Sequencing depth and taxonomic complexity strongly impact binning performance 14.

A benchmarking study of binning tools found that neural network-based tools consistently outperformed others in genome recovery from both real samples and simulated samples with realistic taxonomic complexity, though at higher computational cost 14. By integrating and refining genome bins from the top three binning tools, the study recovered more than 30 percent more high-quality genomes than previous methods 14.

Options and Tradeoffs in Assembly Strategies

Individual Assembly with Gene Clustering

Individual assembly followed by clustering of near-identical genes is preferable when samples are very different or when strain-level resolution is important 6. This approach avoids the risk of mixing data from closely related strains, which can hamper the assembly process 6. However, genes that display too low abundance to be assembled from individual samples may be missed 6.

Mix Assembly

The mix-assembly approach combines individual and co-assembly strategies 6. In a study of the Baltic Sea, this approach resulted in a more extensive nonredundant gene set than the other approaches, with more genes predicted to be complete and functionally annotated 6. The mix assembly consisted of 67 million genes that were functionally and taxonomically annotated 6.

Sequential Co-Assembly

Sequential co-assembly reduces the assembly of duplicate reads through successive application of single-node computing tools for read assembly and mapping 5. This approach shortens assembly time, uses less memory than traditional co-assembly, and produces significantly fewer assembly errors 5. It is particularly useful for very large datasets that cannot be handled by traditional co-assembly 5.

Long-Read and Hybrid Approaches

Long-read sequencing technologies can significantly improve metagenome assemblies. Highly accurate PacBio HiFi reads can yield hundreds of near-complete metagenome-assembled genomes from a single sample 16. The accuracy of Oxford Nanopore Technologies reads has increased to a per-base error rate of 1 to 2 percent 16. The nanoMDBG assembler supports the latest ONT reads through an error correction pre-processing step in minimizer-space 16.

Co-barcoding approaches can also improve assembly quality. The MetaTrass tool uses a strategy of binning first and assembling later to assemble high-quality draft genomes based on metagenomics co-barcoding reads and public reference genomes 17. Applied to single tube long fragment reads datasets for four human faecal samples, MetaTrass generated more high-quality draft genomes with longer contiguity and higher resolution than common combination strategies of genome assembling and binning 17.

Observations and Measurements in Co-Assembly

Sequencing Depth and Coverage

Sequencing depth is a critical factor in co-assembly success. Deeper sequencing provides more coverage for low-abundance organisms, but also increases computational requirements. The relationship between sequencing depth and assembly quality is not linear, and the optimal depth depends on the complexity of the community and the goals of the study.

A study of stable isotope probing metagenomics examined how sequencing depth impacts the detection of isotopically labeled genomes 12. The study used a designed microbiome where the identity of labeled genomes and their level of isotopic enrichment were experimentally controlled 12. This ground truth dataset allowed the researchers to empirically assess the accuracy of different analytical models for identifying active taxa 12.

Sample Number and Binning Performance

The number of samples included in a co-assembly affects binning performance. Multi-sample binning is most effective with about 20 samples 14. Using too few samples reduces the coverage information available for binning, while using too many samples can introduce noise from samples with very different community compositions 14.

Assembly Error Rates

Assembly errors can arise from repetitive regions, strain variation, and sequencing errors. The sequential co-assembly method was shown to produce significantly fewer assembly errors than traditional co-assembly 5. This reduction in errors is attributed to the successive application of single-node computing tools that reduce the assembly of duplicate reads 5.

Records and Documentation for Co-Assembly

Metadata and Sample Information

Accurate metadata is essential for interpreting co-assembly results. Each sample should have associated metadata describing the collection site, collection date, environmental conditions, and any relevant experimental variables. The Genomic Data Sharing Policy from the National Institutes of Health provides guidance on data sharing expectations for genomic research 3.

Assembly Parameters and Versions

Documenting the exact parameters used for assembly is critical for reproducibility. This includes the assembler version, k-mer sizes, minimum contig length thresholds, and any other relevant parameters. The FAIR Guiding Principles emphasize that data should be Findable, Accessible, Interoperable, and Reusable, which applies to both raw reads and assembled contigs 4.

Quality Metrics and Assessment Records

Records of assembly quality metrics should be maintained for each assembly. This includes N50, number of contigs, total assembled bases, and the number of metagenome-assembled genomes recovered. The MAG-E framework provides a systematic approach for evaluating entire metagenome-assembled genome pipelines 15.

Common Failure Patterns in Co-Assembly

Chimeric Assemblies from Strain Mixing

The most common failure pattern in co-assembly is the creation of chimeric contigs from closely related strains. When reads from multiple samples are combined, the assembler may incorrectly join sequences from different strains, producing contigs that do not exist in any single genome. This risk is particularly high when samples contain multiple strains of the same species 6.

Memory Exhaustion on Large Datasets

Traditional co-assembly of large datasets can exhaust available memory. The sequential co-assembly method was developed specifically to address this limitation 5. Researchers working with datasets larger than available memory should consider sequential co-assembly or other memory-efficient approaches.

Reduced Binning Performance with Too Many Samples

Using too many samples in a co-assembly can reduce binning performance 14. The optimal number of samples for multi-sample binning is about 20 14. Researchers should carefully consider which samples to include in a co-assembly, instead of combining all available samples.

Loss of Strain-Level Resolution

Co-assembly can obscure strain-level differences between samples. When the research question requires distinguishing between closely related strains, individual assembly is preferable 6. The mix-assembly approach can partially address this limitation by combining individual and co-assembly results 6.

Quality Controls and Validation

Completeness and Contamination Assessment

Metagenome-assembled genomes should be assessed for completeness and contamination using tools such as CheckM2 or GUNC. A benchmarking study found that CheckM2 systematically overestimates completeness and underestimates contamination, and that this is partially ameliorated when using GUNC 15. Researchers should be aware of these biases when interpreting quality metrics.

Read Mapping Validation

After co-assembly, reads should be mapped back to the assembled contigs to assess how well the assembly represents the original data. Low mapping rates may indicate that the assembly missed significant portions of the community. The nf-UnO pipeline uses read mapping to metagenomically assembled genomes to detect potential etiological agents 11.

Comparison with Reference Genomes

When reference genomes are available, assembled contigs can be compared to references to assess assembly accuracy. This comparison can identify misassemblies and chimeric contigs. The NCBI Data Resources provide access to reference genomes for comparison 2.

Safety and Regulatory Context

Data Sharing and Privacy

Metagenomic data may contain human sequences if host DNA is present in the samples. Researchers should follow applicable data sharing policies, such as the Genomic Data Sharing Policy from the National Institutes of Health 3. This policy provides expectations for data sharing in genomic research 3.

Pathogen Detection and Biosafety

Co-assembly of clinical or outbreak samples may detect pathogenic organisms. The nf-UnO pipeline was developed to identify novel pathogens from metagenomic shotgun sequencing of epidemiologically related foodborne outbreak specimens 11. Researchers working with potentially pathogenic samples should follow appropriate biosafety protocols and consult with relevant authorities if potential pathogens are detected.

Data Management and Reproducibility

The FAIR Guiding Principles provide a framework for data management that supports reproducibility 4. Researchers should deposit raw reads and assembled contigs in public repositories such as the NCBI Sequence Read Archive 2. The EMBL-EBI Training resources provide guidance on data management best practices 1.

Professional Escalation Criteria

Researchers should seek additional expertise or escalate to more experienced colleagues when encountering the following situations:

• Co-assembly produces highly fragmented contigs with low N50 values despite adequate sequencing depth
• Memory exhaustion occurs repeatedly during co-assembly of datasets that should be manageable with available resources
• Binning produces very few metagenome-assembled genomes or genomes with high contamination estimates
• Read mapping rates are consistently below expected thresholds for the community type being studied
• Detection of potential pathogens in clinical or outbreak samples requires consultation with public health authorities
• Results from co-assembly conflict with results from individual assembly in ways that cannot be explained by known tradeoffs

Frequently Asked Questions

What is the difference between co-assembly and individual assembly?

Co-assembly combines reads from multiple samples into a single assembly run, while individual assembly processes each sample separately. Co-assembly can recover genes that are too low in abundance to be assembled from individual samples, but it increases the risk of mixing data from closely related strains 6. Individual assembly preserves strain-level resolution but may miss low-abundance genes 6.

When should I choose co-assembly over individual assembly?

Choose co-assembly when samples share substantial community membership, such as in longitudinal studies, replicate samples, or time series experiments. Co-assembly is also appropriate when the research question targets community-level gene content instead of strain-level differences. For studies focused on strain-level resolution or when samples come from very different environments, individual assembly is preferable 6.

What is the mix-assembly approach?

Mix-assembly combines individual and co-assembly strategies 6. In a study of the Baltic Sea, this approach produced a more extensive nonredundant gene set than either individual assembly or co-assembly alone, with more genes predicted to be complete and functionally annotated 6.

How does sequential co-assembly reduce computational requirements?

Sequential co-assembly reduces the assembly of duplicate reads through successive application of single-node computing tools for read assembly and mapping 5. This approach shortens assembly time, uses less memory than traditional co-assembly, and produces significantly fewer assembly errors 5.

How many samples should I include in a co-assembly?

Multi-sample binning is most effective with about 20 samples 14. Using too few samples reduces the coverage information available for binning, while using too many samples can introduce noise from samples with very different community compositions 14.

What assemblers are recommended for co-assembly?

MEGAHIT and metaSPAdes are commonly used metagenomic assemblers. A benchmarking study found that metaSPAdes consistently outperformed MEGAHIT in terms of recall or completeness 15. However, MEGAHIT is often faster and uses less memory, making it suitable for large datasets or resource-constrained environments.

How do I assess the quality of a co-assembly?

Key quality metrics include N50, number of contigs, total assembled bases, and the number of metagenome-assembled genomes recovered. Reads should be mapped back to the assembled contigs to assess how well the assembly represents the original data. Metagenome-assembled genomes should be assessed for completeness and contamination using tools such as CheckM2 or GUNC 15.

Can co-assembly be used for pathogen detection?

Yes, co-assembly can be used for pathogen detection. The nf-UnO pipeline uses metagenomic co-assembly, multiple binning programs, and read mapping to metagenomically assembled genomes to detect potential etiological agents found in common across outbreak specimens 11. This approach is particularly useful for identifying novel pathogens that may not be detected by reference-based methods.

Related Bioinformatics Guides

• Multi-Omics Integration Strategies
• Long-Read Genome Assembly and Polishing Strategies
• MicroRNA Target Prediction Tools
• Predicting AMR from Genomic Data
• Transcriptome Assembly Without a Reference Genome

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.
• Sequential co-assembly reduces computational resources and errors in metagenome-assembled genomes.. Cell reports methods, 2025.
• Evaluating metagenomic assembly approaches for biome-specific gene catalogues.. Microbiome, 2022.
• Integrative meta-omics in Galaxy and beyond.. Environmental microbiome, 2023.
• Metagenomic Assembly and Gene Prediction.. Methods in molecular biology (Clifton, N.J.), 2026.
• Reference-free structural variant detection in microbiomes via long-read co-assembly graphs.. Bioinformatics (Oxford, England), 2024.
• Carbohydrate-active enzyme annotation in microbiomes using dbCAN.. bioRxiv : the preprint server for biology, 2024.
• nf-UnO pipeline: A metagenomic co-assembly pipeline for novel pathogen detection from mNGS outbreak sets.. Bioinformatics (Oxford, England), 2025.
• A standardized quantitative analysis strategy for stable isotope probing metagenomics.. mSystems, 2023.
• TOFU-MAaPO: fast, scalable and reproducible analysis of large metagenome sequence data from the Sequence Read Archive.. 2026.
• Comprehensive benchmarking of metagenomic binning tools reveals key factors for improved genome recovery.. 2026.
• End-to-end evaluation of pipelines for metagenome-assembled genomes reveals hidden performance gaps. 2026.
• High-quality metagenome assembly from nanopore reads with nanoMDBG.. 2026.
• MetaTrass: High-quality metagenome assembly on the human gut microbiome by co-barcoding sequencing reads. bioRxiv, 2021.
• Metagenomic pipeline for identifying co-infections among distinct SARS-CoV-2 variants of concern: study cases from Alpha to Omicron. Scientific Reports, 2022.
• The Master Database of All Possible RNA Sequences and Its Integration with RNAcmap for RNA Homology Search. bioRxiv, 2023.
• Exploring high-quality microbial genomes by assembling short-reads with long-range connectivity. bioRxiv, 2023.
• Mitochondrial Genome Sequence of the Scabies Mite Provides Insight into the Genetic Diversity of Individual Scabies Infections. PLoS Neglected Tropical Diseases, 2016.
• Recovering genomes from uncultured fungi with single-cell genomics. Journal of Bioscience and Bioengineering, 2026.
• SqueezeMeta, a highly portable, fully automatic metagenomic analysis pipeline. Frontiers in Microbiology, 2019.

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