Evaluating Genome Assembly Quality in the Era of Long Reads

By Dr. Zubair Khalid, DVM, MS, PhD · 2026-08-21

Quick Answer

• Evaluate bacterial genome assemblies with both contiguity metrics like N50 and completeness metrics like BUSCO, since long-read assemblies can be highly contiguous yet still contain errors.
• Use QUAST for assembly statistics and BUSCO for gene completeness assessment, then compare results against application-specific thresholds for pathogen surveillance or comparative genomics.
• No single metric confirms assembly quality, and long-read-only assemblies may require polishing with additional data to reach accuracy comparable to hybrid approaches.

At a Glance

The table below summarizes the primary metrics used to evaluate bacterial genome assemblies, what each metric measures, and the practical questions they answer for researchers working with long-read data.

The Shift Toward Long-Read Bacterial Genome Assembly

Bacterial genome assembly has moved from short-read dominance toward long-read sequencing platforms that produce reads spanning thousands of bases. This shift addresses a fundamental limitation of short-read data. Repetitive regions in bacterial genomes, including insertion sequences, ribosomal RNA operons, and mobile genetic elements, create ambiguity when assembling from fragments that cannot span the repeated region. Long single-molecule sequencing reads resolve many of these repeats because a single read can traverse the entire repetitive element and connect flanking unique sequence. Accurate genome assembly is hampered by repetitive regions, and although long single-molecule sequencing reads are better able to resolve genomic repeats than short-read data, most long-read assembly algorithms do not provide the repeat characterization necessary for producing optimal assemblies (Flye assembly algorithm study).

The practical consequence for researchers is substantial. Assemblies that incorporate long reads are approximately 48 times more contiguous than those built from short reads alone, based on analysis of insect genome assemblies where long-read technology was compared against earlier approaches (insect genome sequencing analysis). While that observation comes from eukaryotic genomes, the principle applies directly to bacterial genomes, where repetitive elements are common and often carry clinically relevant genes.

For bacterial pathogen surveillance, the goal is complete and accurate genome reconstruction. A 2026 evaluation of clinical Enterobacterales isolates compared long-read-only assembly using Oxford Nanopore Technologies against hybrid assembly that combined long and short reads. The study found that long-read-only assembly with appropriate polishing could achieve accuracy comparable to hybrid approaches, with median zero single nucleotide variants and zero insertions or deletions per genome in the best-performing assembly and polishing combination (Nanopore long-read-only assembly study). This finding matters for laboratories considering whether they need to maintain both long-read and short-read platforms or whether long-read-only workflows can support their surveillance objectives.

Core Principles of Assembly Quality Assessment

Contiguity Metrics and Their Interpretation

Contiguity metrics describe how completely a genome is represented as continuous sequence. The N50 statistic is the most widely reported contiguity measure. It represents the length at which half of the total assembly is contained in contigs of that size or larger. A higher N50 indicates fewer gaps and longer continuous sequence blocks.

For bacterial genomes, which typically range from roughly 1 to 10 megabases, a complete assembly should ideally produce a single circular chromosome and one contig per plasmid. The N50 for such an assembly equals the chromosome length. In practice, assemblies fall short of this ideal, and the N50 provides a quick numeric summary of how close the assembly comes to complete representation.

The NGA50 metric extends this concept by measuring contiguity relative to a reference genome. It calculates N50 after breaking contigs at misjoins identified by aligning the assembly to a reference. This metric penalizes assemblies that join sequences incorrectly, since a misjoin that artificially inflates N50 will be broken during the calculation. The Flye assembler benchmark reported that it nearly doubled the NGA50 of the human genome assembly compared with existing assemblers, demonstrating that assembly algorithms differ substantially in their ability to produce both contiguous and correctly ordered sequence (Flye assembly algorithm study).

Completeness Assessment With BUSCO

Contiguity alone does not establish assembly quality. An assembly can be highly contiguous while missing genes or containing errors. Completeness assessment addresses this gap by searching for conserved single-copy orthologs expected in the target taxonomic group.

The Benchmarking Universal Single-Copy Orthologs approach, commonly called BUSCO, compares an assembly against a curated set of genes expected to be present in single copy across a taxonomic lineage. The output reports the percentage of these expected genes found complete, fragmented, or missing. A bacterial assembly with high completeness contains nearly all expected conserved genes, while a low completeness score indicates either sequencing gaps, assembly errors, or contamination that obscured genuine sequence.

The Chinese medaka genome assembly provides an example of BUSCO reporting in practice. That eukaryotic assembly achieved a BUSCO benchmark of 95.1 percent for the genome assembly and 94.6 percent for the gene annotation (Chinese medaka genome assembly study). For bacterial genomes, completeness expectations are typically higher because bacterial genomes are smaller and more gene-dense, but the principle remains the same. Researchers should report BUSCO scores alongside contiguity metrics to give reviewers and collaborators a complete picture of assembly quality.

Base-Level Accuracy and Polishing

Contiguity and completeness describe the structure of an assembly, but they do not guarantee that every base is correct. Base-level accuracy matters for applications that depend on precise sequence, including variant calling, antimicrobial resistance gene detection, and phylogenetic analysis.

Long-read sequencing platforms historically produced higher error rates than short-read platforms. Recent improvements in sequencing chemistry and basecalling have reduced these errors substantially. The 2026 Enterobacterales study used Oxford Nanopore R10.4.1 flow cells with Dorado super-high-accuracy basecalling and found that the best long-read-only assembly and polishing combination achieved median quality values of 100, equivalent to an estimated error rate of one error per 10 billion bases. Only 4 of 92 genome sequences had more than 10 single nucleotide variants or insertions or deletions (Nanopore long-read-only assembly study).

Polishing is the process of correcting residual errors in an assembly after the initial assembly step. The Enterobacterales study compared several polishing approaches. Medaka polishing with unsubsampled long reads produced small improvements in insertions and deletions but did not improve single nucleotide variant accuracy for either Flye or Autocycler assemblies. This finding suggests that for modern high-accuracy long-read data, the primary assembly step determines single nucleotide accuracy, while polishing mainly corrects residual indel errors (Nanopore long-read-only assembly study).

Practical Workflow for Assembly Quality Assessment

Step 1: Generate the Assembly

Select an assembler appropriate for your sequencing platform and data characteristics. The 2026 Enterobacterales evaluation compared Flye, Hybracter long, and Autocycler for long-read-only assembly, alongside hybrid assemblers including Hybracter hybrid and Unicycler. In that study, Autocycler circularized the most chromosomes at 87 of 92 genomes, while Flye performed worse than other assemblers on almost all metrics (Nanopore long-read-only assembly study).

This result does not mean Flye is universally unsuitable. The Flye algorithm was designed to handle repetitive regions by constructing a repeat graph from error-prone long reads, and it has demonstrated strong performance in other contexts (Flye assembly algorithm study). The practical lesson is that assembler performance varies by dataset, and laboratories should evaluate multiple assemblers on their own data instead of assuming one tool will always perform best.

Step 2: Calculate Assembly Statistics

Run QUAST or an equivalent tool to generate summary statistics for each assembly. Report the number of contigs, total assembly length, N50, and the largest contig. For bacterial genomes, also examine whether the assembly contains circular contigs, which indicate complete chromosome or plasmid reconstruction.

Compare these statistics against expectations for your organism. A complete bacterial chromosome should assemble as a single circular contig. Plasmids should appear as additional circular contigs. If the assembly contains many small contigs, this may indicate incomplete assembly, contamination, or the presence of multiple plasmids that were not fully resolved.

Step 3: Assess Completeness With BUSCO

Run BUSCO with the appropriate lineage dataset for your organism. For bacteria, use the bacteria lineage or a more specific lineage if one is available for your taxonomic group. Record the percentage of complete, fragmented, and missing BUSCO genes.

A high-quality bacterial assembly should have greater than 95 percent complete BUSCO genes. Lower completeness may indicate sequencing depth issues, assembly errors, or contamination. If completeness is low, consider whether additional sequencing or a different assembler might improve the result.

Step 4: Evaluate Base-Level Accuracy

For applications that require precise sequence, assess base-level accuracy by comparing the assembly against a trusted reference or by examining variant calls. The Enterobacterales study used short-read data to evaluate long-read-only assemblies, treating the hybrid assembly as a benchmark for comparison (Nanopore long-read-only assembly study).

If you have short-read data available, consider polishing the long-read assembly with short reads. Tools such as Polypolish and Pypolca were evaluated in the Enterobacterales study and can correct residual errors. If you do not have short-read data, polishing with unsubsampled long reads using Medaka can improve indel accuracy, though it may not correct single nucleotide variants (Nanopore long-read-only assembly study).

Step 5: Check Plasmid Reconstruction

Plasmids carry clinically important genes, including antimicrobial resistance determinants. Their reconstruction is therefore essential for pathogen surveillance. The Enterobacterales study found that plasmid sequence reconstruction was comparable among all assemblers except Flye, with all assemblers recovering 90 to 96 percent of plasmids (Nanopore long-read-only assembly study).

Assess plasmid recovery by examining circular contigs smaller than the chromosome and by comparing plasmid content against expectations for your organism. If plasmids are missing or fragmented, consider whether the assembly approach adequately represents extrachromosomal elements.

Options and Tradeoffs in Assembly Strategies

Long-Read-Only Assembly

Long-read-only assembly offers lower cost and higher throughput compared with hybrid approaches because it requires only a single sequencing platform. The Enterobacterales study evaluated this approach as a potential alternative for national pathogen surveillance programs. The best-performing long-read-only combination, Autocycler with Medaka polishing using unsubsampled long reads, achieved accuracy comparable to hybrid assemblies (Nanopore long-read-only assembly study).

The tradeoff is that long-read-only assembly depends heavily on the accuracy of the sequencing platform and basecalling pipeline. Laboratories using older flow cells or less accurate basecalling may see higher error rates. The study specifically used R10.4.1 flow cells with Dorado super-high-accuracy basecalling, and results may differ with other configurations (Nanopore long-read-only assembly study).

Hybrid Assembly

Hybrid assembly combines long reads for contiguity with short reads for accuracy. This approach has been the standard for high-quality bacterial genome assembly because it leverages the strengths of both platforms. The Enterobacterales study included hybrid assemblers as comparators and found that the best long-read-only approach matched hybrid accuracy (Nanopore long-read-only assembly study).

The tradeoff is cost and complexity. Hybrid assembly requires access to both long-read and short-read sequencing platforms, increasing equipment and reagent costs. It also requires more bioinformatics expertise to integrate data from multiple platforms.

Polishing Strategies

Polishing is not a single step but a family of approaches with different strengths. The Enterobacterales study compared Medaka polishing with subsampled and unsubsampled long reads, as well as Polypolish and Pypolca with short reads. Medaka with unsubsampled long reads produced small improvements in indels but not single nucleotide variants for both Flye and Autocycler assemblies (Nanopore long-read-only assembly study).

The practical implication is that polishing should be matched to the error profile of your data. If your assembly has residual indel errors, Medaka with full long-read data may help. If single nucleotide variants are the concern, short-read polishing may be necessary, or the basecalling pipeline may need adjustment.

Records and Measurements for Assembly Quality

What to Record

Maintain a record of assembly statistics for every genome assembled in your laboratory. At minimum, record the assembler and version, sequencing platform and chemistry, basecalling software and version, polishing tools and parameters, number of contigs, total assembly length, N50, largest contig, BUSCO completeness percentage, and any quality value or error estimates.

This record serves multiple purposes. It allows you to track whether assembly quality changes over time as sequencing platforms and software improve. It provides documentation for publications and data submissions. It also enables troubleshooting when an assembly fails quality thresholds.

Quality Thresholds by Application

Different applications require different levels of assembly quality. For routine species identification, a fragmented assembly with high completeness may be sufficient. For antimicrobial resistance gene detection, base-level accuracy becomes critical because a single nucleotide error can alter a resistance gene prediction. For phylogenetic analysis, both contiguity and accuracy matter because missing or misassembled regions can distort evolutionary relationships.

The Enterobacterales study provides a benchmark for surveillance applications. The best-performing assembly and polishing combination achieved median zero single nucleotide variants and zero insertions or deletions per genome, with a median quality value of 100 (Nanopore long-read-only assembly study). Laboratories conducting pathogen surveillance should aim for similar accuracy, recognizing that this level of quality requires modern sequencing chemistry and careful polishing.

The table below summarizes recommended quality targets for common bacterial genomics applications.

When to Escalate

If an assembly fails to meet quality thresholds, do not simply accept the result. Investigate the cause before proceeding with downstream analysis. Common causes of poor assembly quality include insufficient sequencing depth, contaminated DNA preparations, mixed cultures, and assembler parameter mismatches.

Escalate to a supervisor or bioinformatics specialist if you cannot resolve assembly quality issues after troubleshooting. Poor assemblies can lead to incorrect biological conclusions, including missed resistance genes, false variant calls, and incorrect phylogenetic placement. The cost of re-sequencing is often lower than the cost of publishing or acting on incorrect results.

Common Failure Patterns in Long-Read Assembly

Fragmented Assemblies Despite Long Reads

Long reads should produce highly contiguous assemblies, but fragmentation can still occur. This pattern often indicates insufficient sequencing depth, where coverage gaps prevent the assembler from connecting contigs. It can also result from high levels of repetitive sequence that confuse the assembler even with long reads.

Check sequencing depth first. Bacterial genome assembly typically requires at least 50-fold coverage, and higher coverage may be needed for genomes with extensive repetitive content. If depth is adequate, consider whether the assembler parameters are appropriate for your data.

High Completeness With Low Base Accuracy

An assembly can contain all expected genes while still having errors at the base level. This pattern is particularly problematic because it may not be detected by BUSCO or contiguity metrics alone. The Enterobacterales study found that polishing with unsubsampled long reads improved indel accuracy but not single nucleotide variant accuracy, suggesting that some errors originate during assembly instead of from sequencing noise (Nanopore long-read-only assembly study).

If you suspect base-level errors, compare your assembly against a trusted reference or use short-read data to validate. The absence of short-read data makes this validation difficult, which is a limitation of long-read-only workflows.

Plasmid Loss or Fragmentation

Plasmids are often present in multiple copies and can be difficult to assemble, particularly if they share sequence with the chromosome or with other plasmids. The Enterobacterales study found that plasmid reconstruction was comparable among most assemblers but that Flye performed worse on this metric (Nanopore long-read-only assembly study).

If plasmids are missing from your assembly, consider whether the DNA extraction method preserved extrachromosomal elements. Some extraction methods shear plasmid DNA, making it difficult to assemble. Consider whether the assembler you selected handles plasmids well, and evaluate alternative assemblers if plasmid recovery is consistently poor.

Contamination and Mixed Cultures

Contamination introduces sequence from other organisms into the assembly, inflating the total length and reducing BUSCO completeness. Mixed cultures produce the same effect because the assembler attempts to assemble multiple genomes simultaneously.

Check for contamination by examining the assembly for unexpected genes or by comparing GC content against expectations for your organism. If contamination is suspected, return to the original culture and confirm purity before repeating the sequencing and assembly.

Safety and Regulatory Context for Genome Assembly Work

Biosafety Considerations

Genome assembly work begins with DNA extraction from cultured organisms. The biosafety level required depends on the organism being studied. The CDC and NIH Biosafety in Microbiological and Biomedical Laboratories provides guidance on risk assessment, containment, and decontamination for work with infectious agents. Consult this guidance before beginning work with any organism that may pose a risk to laboratory personnel.

Standard precautions for bacterial genome assembly include working in a biosafety cabinet when handling cultures, using appropriate personal protective equipment, and decontaminating work surfaces after each session. These practices protect both the researcher and the integrity of the sample, since contamination from the environment can compromise sequencing results.

Antimicrobial Resistance Context

Genome assembly of clinical isolates often focuses on antimicrobial resistance genes. The World Health Organization defines antimicrobial resistance as a global public health threat and emphasizes the importance of surveillance and stewardship. Genome-based surveillance contributes to this effort by enabling rapid detection of resistance determinants and tracking of resistant clones.

When assembling genomes for antimicrobial resistance surveillance, recognize that the accuracy of resistance gene prediction depends on assembly quality. A single nucleotide error can alter a resistance gene prediction, leading to incorrect clinical or public health conclusions. The Clinical and Laboratory Standards Institute provides standards for antimicrobial susceptibility testing that complement genomic approaches, and laboratories should understand the relationship between genotypic predictions and phenotypic susceptibility results.

Quality Control and Authenticated Materials

The American Type Culture Collection provides authenticated biological materials for research and quality control. Using authenticated reference strains in your assembly workflow provides a benchmark for evaluating assembly quality. If your assembly of a reference strain does not match the expected genome, this indicates a problem with your sequencing or assembly pipeline.

The FDA Bacteriological Analytical Manual provides official methods for food microbiology that include quality control principles applicable to genome assembly. While the manual focuses on foodborne pathogen detection, its emphasis on validated methods and quality control aligns with best practices for genome assembly.

Evidence Limitations and Interpretation

The Rapidly Changing Technology Landscape

Long-read sequencing technology is improving rapidly. The Enterobacterales study used R10.4.1 flow cells with Dorado v5.0.0 super-high-accuracy basecalling, which represents a specific point in the technology trajectory (Nanopore long-read-only assembly study). Newer chemistry and basecalling software may produce different error profiles, and results from this study may not transfer directly to other configurations.

Researchers should validate their own pipelines instead of assuming that published benchmarks apply to their specific setup. This validation should include sequencing a well-characterized reference strain and comparing the assembly against the expected genome.

The Challenge of Ground Truth

The Enterobacterales study noted that the ground truth for plasmid reconstruction was uncertain (Nanopore long-read-only assembly study). Plasmids are difficult to validate because they are often present in multiple copies and may be rearranged relative to reference sequences. This uncertainty affects all plasmid assessments, beyond those in the study.

When evaluating plasmid reconstruction, recognize that the absence of a plasmid in an assembly does not prove the plasmid was absent from the original isolate. It may indicate that the plasmid was not fully assembled or that it was lost during culture or DNA extraction.

Taxonomic Variation in Assembly Difficulty

Bacterial genomes vary substantially in their assembly difficulty. Genomes with high GC content, extensive repetitive sequence, or multiple nearly identical plasmids present greater challenges than genomes with simpler structures. The Enterobacterales study focused on a specific taxonomic group, and results may not generalize to all bacteria (Nanopore long-read-only assembly study).

For unusual or difficult organisms, expect that assembly may require additional optimization. Consider testing multiple assemblers and polishing strategies, and be prepared to invest more time in achieving acceptable quality.

Professional Escalation Criteria

When to Seek Additional Expertise

Genome assembly troubleshooting can consume substantial time and resources. Escalate to a bioinformatics specialist or supervisor when you encounter any of the following situations:

• Assembly quality does not improve after testing multiple assemblers and polishing strategies
• BUSCO completeness remains below 90 percent despite adequate sequencing depth
• Base-level accuracy cannot be validated because no reference or short-read data are available
• Plasmid reconstruction consistently fails for a particular organism or sample type
• Contamination cannot be resolved by repeating DNA extraction and sequencing

When to Consider Re-Sequencing

Re-sequencing is appropriate when the existing data cannot produce an acceptable assembly. This decision should be based on the quality metrics described above and the requirements of the downstream application. For pathogen surveillance, the cost of re-sequencing is justified when assembly quality falls below the accuracy achieved in published benchmarks.

Before re-sequencing, review the entire workflow from culture through assembly. A problem at any step can compromise the final result. Confirm culture purity, assess DNA quality and quantity, verify sequencing depth, and check that the assembler and polishing parameters match the data characteristics.

A Decision Framework for Matching Assembly Quality to Downstream Application

Defining the Quality Threshold Problem

Researchers often generate a bacterial genome assembly and then face a more difficult question than how to measure quality. That question is whether the measured quality is sufficient for the intended use. A genome assembled for species identification does not require the same accuracy as a genome assembled for antimicrobial resistance gene surveillance. A genome used for phylogenetic placement may tolerate different error profiles than one used for variant-level outbreak tracking. The practical problem is that many laboratories apply a single quality standard to all assemblies regardless of the downstream application, which leads either to wasted sequencing effort on overvalidated assemblies or to incorrect conclusions from undervalidated ones.

This section provides a structured decision framework that connects assembly quality metrics to specific downstream applications. The framework uses the evidence from the 2026 Enterobacterales study as a benchmark for what modern long-read-only assembly can achieve, then maps those capabilities to the quality demands of common research and surveillance tasks (Nanopore long-read-only assembly study). The goal is to help researchers decide before they assemble what quality level they actually need, so they can choose the appropriate sequencing depth, assembler, and polishing strategy from the start.

Application Categories and Their Quality Demands

Species Identification and Taxonomic Assignment

Species identification from genome assemblies relies primarily on gene content and overall genome structure instead of on base-level precision. The presence of conserved marker genes, the overall genome size, and the GC content provide enough information to assign a genome to a species or genus. For this application, BUSCO completeness is the most important metric because a fragmented assembly that misses conserved genes can lead to incorrect taxonomic conclusions.

A BUSCO completeness score above 95 percent with the appropriate lineage dataset is generally sufficient for species identification. The Chinese medaka genome assembly achieved 95.1 percent BUSCO completeness and was considered high quality for its purpose, demonstrating that this threshold represents a reasonable standard across taxonomic groups (Chinese medaka genome assembly study). For bacterial genomes, which are smaller and more gene-dense than eukaryotic genomes, completeness expectations should be higher, and scores below 95 percent should trigger investigation.

Base-level accuracy matters less for species identification because taxonomic assignment depends on gene presence and sequence similarity at the level of conserved markers instead of on every nucleotide being correct. A small number of single nucleotide variants or insertions or deletions in non-marker regions will not change a species assignment. This means that laboratories performing only species identification can accept assemblies that would be inadequate for resistance gene surveillance.

Antimicrobial Resistance Gene Detection

Antimicrobial resistance gene detection requires base-level accuracy because resistance determinants are identified by sequence similarity to known resistance genes. A single nucleotide error can alter a resistance gene prediction, either creating a false positive by making a non-resistance gene look like a resistance gene or creating a false negative by disrupting a genuine resistance determinant. The World Health Organization defines antimicrobial resistance as a global public health threat and emphasizes the importance of surveillance, which makes accurate resistance gene detection a public health priority.

For this application, the quality target should be zero single nucleotide variants in resistance genes. The 2026 Enterobacterales study provides a benchmark for what is achievable with modern long-read-only assembly. The best-performing combination of Autocycler with Medaka polishing using unsubsampled long reads achieved median zero single nucleotide variants and zero insertions or deletions per genome, with a median quality value of 100 (Nanopore long-read-only assembly study). Laboratories conducting resistance gene surveillance should aim for this level of accuracy.

The practical implication is that resistance gene detection requires more than a high BUSCO score. A genome can be complete in terms of gene content while still containing errors in the specific genes that matter for resistance prediction. Researchers should validate resistance gene calls by examining the underlying sequence alignments instead of relying solely on automated annotation.

Phylogenetic and Evolutionary Analysis

Phylogenetic analysis depends on both contiguity and base-level accuracy because missing or misassembled regions can distort evolutionary relationships. A fragmented assembly that breaks a conserved gene into multiple contigs may cause the phylogenetic reconstruction to treat the fragments as separate loci. A misassembled region that joins sequences from different parts of the genome can create false phylogenetic signal.

For phylogenetic applications, the N50 metric matters because it indicates whether genes and operons are represented as continuous sequence. The observation that assemblies incorporating long reads are approximately 48 times more contiguous than those built from short reads alone (insect genome sequencing analysis) is directly relevant here, because phylogenetic analysis benefits from the improved contiguity that long reads provide.

The quality target for phylogenetic analysis is a complete circular chromosome with zero indels in conserved regions. This is a higher standard than species identification but potentially lower than resistance gene surveillance, depending on the phylogenetic question. For deep evolutionary relationships, base-level accuracy in conserved genes is critical. For strain-level phylogenetics, the focus shifts to variant detection across the genome, which requires the same accuracy as resistance gene surveillance.

Pathogen Surveillance and Outbreak Tracking

Pathogen surveillance combines elements of all the applications above. It requires accurate species identification, reliable resistance gene detection, and sufficient phylogenetic resolution to track transmission chains. The NCBI Pathogen Detection system exemplifies this application by integrating genome assemblies from public health laboratories to identify clusters of related pathogens.

For surveillance applications, the quality target should match the published benchmarks from the Enterobacterales study. That study was conducted under a national surveillance program in England and evaluated long-read-only assembly as a potential alternative to hybrid assembly for routine surveillance (Nanopore long-read-only assembly study). The finding that long-read-only assembly could match hybrid accuracy suggests that surveillance programs can consider single-platform workflows, but only if they achieve the accuracy demonstrated in that study.

The practical implication is that surveillance laboratories should establish internal quality thresholds based on published benchmarks and reject assemblies that fall below those thresholds. This requires a documented quality control process that includes reference strains and regular validation of the assembly pipeline.

A Structured Decision Procedure

Step 1: Define the Primary Application Before Assembly

The first step in the decision framework is to define the primary application before generating the assembly. This decision determines the quality targets, the sequencing depth, the assembler choice, and the polishing strategy. A laboratory that defines its application as resistance gene surveillance will make different choices than one focused on species identification.

Document the application in the laboratory record alongside the sample information. This documentation ensures that quality decisions are traceable and that downstream analysts understand the limitations of the assembly.

Step 2: Select Quality Targets Based on Application

Use the application categories above to select specific quality targets. The table below summarizes the recommended targets for each application category.

These targets are starting points instead of universal standards. Different organisms and different research questions may require different thresholds. The key is to define targets before assembly and document the rationale.

Step 3: Choose Sequencing and Assembly Parameters Based on Targets

The quality targets selected in Step 2 should inform the sequencing and assembly parameters. For species identification, lower sequencing depth may be acceptable because base-level accuracy is less critical. For resistance gene surveillance, the highest available sequencing accuracy should be used, and polishing should be planned from the start.

The Enterobacterales study used Oxford Nanopore R10.4.1 flow cells with Dorado v5.0.0 super-high-accuracy basecalling, which represents a specific configuration that achieved high accuracy (Nanopore long-read-only assembly study). Laboratories using older flow cells or less accurate basecalling should expect lower accuracy and should either adjust their quality targets or invest in updated technology.

Step 4: Evaluate the Assembly Against Targets

After assembly and polishing, evaluate the assembly against the targets defined in Step 2. Use QUAST for contiguity statistics, BUSCO for completeness, and variant calling against a trusted reference for base-level accuracy. Record all metrics in the laboratory record.

If the assembly meets all targets, proceed with downstream analysis. If it fails any target, do not proceed. Investigate the cause and either improve the assembly or re-sequence.

Step 5: Document and Review Quality Decisions

Document the quality assessment for every assembly, including the metrics, the targets, and the decision to accept or reject. This documentation supports publications, data submissions, and future troubleshooting. It also provides a basis for reviewing whether the quality targets are appropriate for the applications being performed.

Review quality decisions periodically, especially when sequencing technology or assembly software changes. The rapid improvement in long-read sequencing means that targets that were difficult to achieve last year may be routine this year, and targets that seemed adequate may no longer be sufficient for new applications.

A Record System for Assembly Quality Decisions

What to Record for Each Assembly

Maintain a structured record for every genome assembly that includes the application definition, the quality targets, the assembly parameters, and the quality metrics. The record should include the following fields:

• Sample identifier and source
• Primary application and expected downstream use
• Sequencing platform, chemistry, and basecalling software versions
• Assembler name and version
• Polishing tools and parameters
• Number of contigs
• Total assembly length
• N50 and largest contig
• BUSCO completeness percentage with lineage dataset
• Base-level accuracy metrics including single nucleotide variants and insertions or deletions
• Plasmid reconstruction assessment
• Quality targets and whether they were met
• Decision to accept or reject the assembly
• Any troubleshooting actions taken

This record serves multiple purposes. It allows comparison across assemblies to identify trends in quality. It provides documentation for publications and data submissions. It enables troubleshooting when an assembly fails quality thresholds. It also supports training by giving new researchers examples of quality assessment in practice.

Using Records to Identify Systematic Problems

Beyond documenting individual assemblies, the record system enables identification of systematic problems. If multiple assemblies from the same sequencing run fail quality targets, the problem likely lies in the sequencing run instead of in individual samples. If assemblies from a particular organism consistently fail plasmid reconstruction, the problem may lie in the DNA extraction method instead of in the assembler.

Review the record system regularly to identify patterns. The American Society for Microbiology provides professional education and laboratory practice resources that can help laboratories develop and maintain effective quality systems. The FDA Bacteriological Analytical Manual also emphasizes the importance of validated methods and quality control in food microbiology, principles that apply directly to genome assembly workflows.

Troubleshooting Quality Failures by Application

Species Identification Failures

If BUSCO completeness is below the target for species identification, the most likely causes are insufficient sequencing depth, contamination, or mixed cultures. Check sequencing depth first and confirm that the DNA preparation was pure. The CDC and NIH Biosafety in Microbiological and Biomedical Laboratories provides guidance on laboratory practices that prevent contamination, which is relevant here because contamination from the environment or from other samples can compromise assembly quality.

If completeness remains low after confirming depth and purity, consider whether the BUSCO lineage dataset is appropriate for the organism. Using the wrong lineage dataset can produce misleading completeness scores.

Resistance Gene Detection Failures

If resistance gene detection produces unexpected results, first verify that the assembly meets the base-level accuracy target. The Enterobacterales study found that polishing with unsubsampled long reads improved indel accuracy but not single nucleotide variant accuracy (Nanopore long-read-only assembly study). This means that single nucleotide variants may persist even after polishing and can affect resistance gene predictions.

If the assembly meets accuracy targets but resistance gene predictions still seem incorrect, compare the genomic prediction against phenotypic susceptibility testing. The Clinical and Laboratory Standards Institute provides standards for antimicrobial susceptibility testing that complement genomic approaches. Discrepancies between genotype and phenotype should be investigated instead of assumed to reflect assembly error.

Phylogenetic Analysis Failures

If phylogenetic analysis produces unexpected relationships, examine the assembly for misassembled regions. The NGA50 metric, which breaks contigs at misjoins identified by alignment to a reference, provides a way to detect misassembly. The Flye assembler benchmark reported that it nearly doubled the NGA50 of the human genome assembly compared with existing assemblers (Flye assembly algorithm study), demonstrating that assembler choice affects also contiguity but also correctness.

If misassembly is suspected, consider whether the assembler choice is appropriate for the organism. The Enterobacterales study found that assembler performance varied, with Autocycler circularizing the most chromosomes and Flye performing worse on almost all metrics (Nanopore long-read-only assembly study). Testing multiple assemblers may be necessary for difficult genomes.

Surveillance Program Failures

If a surveillance program consistently produces assemblies that fail quality targets, the problem may be systemic instead of sample-specific. Review the entire workflow from culture through assembly, including DNA extraction, sequencing, basecalling, assembly, and polishing. The American Type Culture Collection provides authenticated reference strains that can be used to validate the workflow. If the assembly of a reference strain does not match the expected genome, this indicates a problem in the workflow instead of in the samples.

Consider whether the quality targets are realistic for the technology being used. The Enterobacterales study achieved high accuracy with specific technology and parameters (Nanopore long-read-only assembly study). Laboratories using different technology may need to adjust targets or upgrade their workflows.

Escalation Criteria Specific to Application

When to Escalate for Species Identification

Escalate to a supervisor or bioinformatics specialist when BUSCO completeness remains below 90 percent after confirming sequencing depth and culture purity. Low completeness may indicate contamination that cannot be resolved by simple troubleshooting, or it may indicate an organism with unusual genomic features that require specialized assembly approaches.

When to Escalate for Resistance Gene Detection

Escalate when resistance gene predictions cannot be validated by phenotypic testing or when base-level accuracy cannot be confirmed because no reference or short-read data are available. The absence of validation data makes it impossible to determine whether resistance gene predictions are correct, and proceeding with unvalidated predictions could lead to incorrect public health conclusions.

When to Escalate for Phylogenetic Analysis

Escalate when phylogenetic results are unstable across different assembly approaches or when misassembly is suspected but cannot be confirmed. Unstable phylogenetic results may indicate that the assembly quality is insufficient for the intended analysis, and additional sequencing or alternative assembly strategies may be needed.

When to Escalate for Surveillance Programs

Escalate when a surveillance program consistently fails quality targets across multiple samples or when the workflow cannot be validated using authenticated reference strains. The NCBI Pathogen Detection system represents a national-level surveillance effort, and laboratories contributing to such efforts must meet quality standards to ensure that public health conclusions are reliable.

The Cost-Benefit Logic of Quality Decisions

The decision framework presented here rests on a simple cost-benefit logic. Higher quality assemblies cost more in sequencing, computing, and personnel time. Lower quality assemblies risk incorrect conclusions. The optimal quality level depends on the cost of being wrong.

For species identification, the cost of being wrong is relatively low because taxonomic assignments are robust to small numbers of errors. For resistance gene surveillance, the cost of being wrong is high because a missed resistance gene can lead to inappropriate treatment or ineffective public health intervention. For phylogenetic analysis, the cost of being wrong depends on the question being asked, with deep evolutionary questions being more sensitive to assembly errors than strain-level questions.

The World Health Organization emphasizes the importance of surveillance for antimicrobial resistance, which means that the cost of incorrect resistance gene detection extends beyond the individual laboratory to public health systems. Laboratories conducting resistance surveillance should therefore invest in the highest quality assembly achievable with their resources.

Limitations of the Decision Framework

The decision framework presented here has several limitations that researchers should recognize. First, the quality targets are based on a limited set of published studies, primarily the 2026 Enterobacterales evaluation (Nanopore long-read-only assembly study) and the insect genome sequencing analysis (insect genome sequencing analysis). These studies represent specific organisms, technologies, and time points, and their results may not generalize to all bacterial species or to future sequencing platforms.

Second, the framework assumes that quality metrics accurately reflect the biological properties that matter for downstream applications. This assumption may not always hold. For example, BUSCO completeness measures the presence of conserved genes but does not guarantee that those genes are correctly assembled or that their regulatory regions are intact.

Third, the framework does not account for organism-specific challenges. Genomes with high GC content, extensive repetitive sequence, or multiple nearly identical plasmids present greater assembly challenges than simpler genomes. The Enterobacterales study focused on a specific taxonomic group, and results may not generalize to all bacteria (Nanopore long-read-only assembly study).

Despite these limitations, the framework provides a practical starting point for laboratories that need to make quality decisions about genome assemblies. The key principle is to define quality targets before assembly, document the rationale, and escalate when targets cannot be met. This principle applies regardless of the specific technology or organism.

Frequently Asked Questions

What is the difference between N50 and BUSCO completeness?

N50 measures contiguity by describing the length at which half of the assembly is contained in contigs of that size or larger. BUSCO completeness measures gene content by searching for conserved single-copy orthologs expected in the target taxonomic group. Both metrics are necessary because an assembly can be highly contiguous while missing genes, or complete while fragmented.

Why is base-level accuracy important for bacterial genome assembly?

Base-level accuracy matters for applications that depend on precise sequence, including variant calling, antimicrobial resistance gene detection, and phylogenetic analysis. A single nucleotide error can alter a resistance gene prediction or create a false variant call. The Enterobacterales study found that the best long-read-only assembly and polishing combination achieved median zero single nucleotide variants per genome, establishing a benchmark for surveillance applications (Nanopore long-read-only assembly study).

How does long-read-only assembly compare with hybrid assembly for bacteria?

The 2026 Enterobacterales study found that the best long-read-only assembly and polishing combination, Autocycler with Medaka polishing using unsubsampled long reads, achieved accuracy comparable to hybrid assemblies (Nanopore long-read-only assembly study). Long-read-only assembly offers lower cost and higher throughput because it requires only one sequencing platform. Hybrid assembly remains valuable when short-read data are already available or when maximum accuracy is required.

What polishing strategy should I use for long-read assemblies?

The Enterobacterales study found that Medaka polishing with unsubsampled long reads produced small improvements in insertions and deletions but not single nucleotide variants for both Flye and Autocycler assemblies (Nanopore long-read-only assembly study). Short-read polishing with tools such as Polypolish and Pypolca can correct errors that long-read polishing misses. The best strategy depends on your data and the error profile of your assembly.

Why are plasmids difficult to assemble?

Plasmids are often present in multiple copies and may share sequence with the chromosome or with other plasmids. They can be sheared during DNA extraction, making them difficult to assemble. The Enterobacterales study found that plasmid reconstruction was comparable among most assemblers but that Flye performed worse on this metric (Nanopore long-read-only assembly study). If plasmids are missing from your assembly, consider whether the DNA extraction method preserved extrachromosomal elements.

How do I know if my assembly is contaminated?

Contamination introduces sequence from other organisms into the assembly, inflating the total length and reducing BUSCO completeness. Check for contamination by examining the assembly for unexpected genes or by comparing GC content against expectations for your organism. If contamination is suspected, return to the original culture and confirm purity before repeating the sequencing and assembly.

What sequencing depth is needed for bacterial genome assembly?

Bacterial genome assembly typically requires at least 50-fold coverage, and higher coverage may be needed for genomes with extensive repetitive content. The Enterobacterales study did not specify a minimum depth, but the high accuracy achieved in that study reflects the use of modern sequencing chemistry and careful polishing (Nanopore long-read-only assembly study). If your assembly is fragmented, check whether sequencing depth is adequate before troubleshooting other causes.

When should I escalate assembly quality issues to a specialist?

Escalate when assembly quality does not improve after testing multiple assemblers and polishing strategies, when BUSCO completeness remains below 90 percent despite adequate sequencing depth, when base-level accuracy cannot be validated, when plasmid reconstruction consistently fails, or when contamination cannot be resolved. Poor assemblies can lead to incorrect biological conclusions, so do not proceed with downstream analysis until quality issues are resolved.

Related Bacteriology Guides

• Evaluating Genome Assembly Quality: Metrics and Tools
• Bovine Tuberculosis: Diagnostic Tools and Wildlife Reservoirs
• Tuberculosis in Wildlife: Mycobacterium bovis Surveillance and Diagnostic Tools
• Which Bacteria Are Common to Raw Poultry? A Safety and Pathogen Guide
• Fowl Cholera: Etiology, Clinical Manifestations, and Control in Poultry

References and Further Reading

• Biosafety in Microbiological and Biomedical Laboratories. CDC and NIH.
• Microbiology Resources. American Society for Microbiology.
• Antimicrobial Resistance. World Health Organization.
• Bacteriological Analytical Manual. U.S. Food and Drug Administration.
• Antimicrobial Susceptibility Testing. Clinical and Laboratory Standards Institute.
• Culture Collections. American Type Culture Collection.
• NCBI Pathogen Detection. National Center for Biotechnology Information.
• Assembly of long, error-prone reads using repeat graphs.. Nature biotechnology, 2019.
• Long Reads Are Revolutionizing 20 Years of Insect Genome Sequencing.. Genome biology and evolution, 2021.
• Nanopore long-read-only genome assembly of clinical Enterobacterales isolates is complete and accurate.. Microbial genomics, 2026.
• A high-quality chromosome-level genome assembly of the Chinese medaka Oryzias sinensis.. Scientific data, 2024.

This article is educational and does not replace validated laboratory procedures, clinical diagnosis, institutional biosafety review, or public-health guidance.