Genome Assembly Quality Assessment: Interpreting Contiguity and Completeness
If you have assembled a genome from sequencing reads, you need to know whether that assembly is reliable enough to answer your biological question. The answer is not a single number but a combination of metrics that measure contiguity, completeness, contamination, and structural correctness. This guide is for bioinformaticians, genomics researchers, and students who need to evaluate a draft or finished genome assembly before publishing, depositing in a repository, or using it for downstream analysis.
At a Glance: Key Quality Metrics
| Metric | What it measures | What it does not measure | Typical target for a good assembly |
|---|---|---|---|
| N50 | Contiguity: half the assembly is in contigs or scaffolds of at least this length | Base-level accuracy, gene content, or ordering of contigs | Bacterial: >100 kbp, Eukaryotic: >1 Mbp (scaffold N50) |
| Coverage (depth) | Average number of reads covering each base | Uniformity across the genome or presence of repeats | 30x to 100x for most projects, higher for polyploid or highly repetitive |
| BUSCO completeness | Fraction of conserved single-copy orthologs present in the assembly | Strain or species specific genes, or de novo gene content | >90% for eukaryotic, >95% for bacterial |
| Contamination | Presence of foreign DNA sequences (e.g., from other organisms, vectors) | Whether contamination is biologically irrelevant (e.g., symbionts) | Minimal: <5% of total assembly length |
| Structural correctness | Whether large scale rearrangements match the true genome | Local base accuracy or small indels | Validated by genetic maps, HiC, or optical maps |
What These Metrics Can and Cannot Show
N50 and Other Contiguity Metrics
N50 is the most cited contiguity statistic. It tells you that at least 50% of the assembled bases lie in contigs or scaffolds of length N50 or longer. Higher N50 values suggest fewer gaps and longer contiguous sequences. Complementary metrics include L50 (the minimum number of contigs needed to reach half the assembly) and NG50 (N50 normalized by estimated genome size). These numbers do not reflect base accuracy, completeness, or ordering. A high N50 can come from a chimeric assembly that incorrectly joins unrelated regions. The EMBL EBI training resources emphasize that contiguity is a necessary but never sufficient condition for quality [2].
Coverage and Its Relationship to Accuracy
Coverage is the average number of reads that align to each base in the assembly. Higher coverage generally improves base calling accuracy, especially with Illumina short reads. However, coverage alone cannot detect systematic errors in repetitive regions or structural variants. Very high coverage from a single sequencing platform (for example, Oxford Nanopore in the ONT only assembly of a Korean male individual [6]) may still leave poly A/T stretches or segmental duplications unresolved. Coverage should be assessed per chromosome or per window to identify outliers that indicate collapsed repeats or contaminants.
Completeness with BUSCO
BUSCO (Benchmarking Universal Single Copy Orthologs) measures how many core genes expected for a given lineage are present in the assembly. A score above 90% (eukaryotic) or 95% (bacterial) is a strong indicator that most coding regions are captured. It does not verify the order of those genes or the integrity of noncoding regions. The Galaxy Training Network provides workflows that run BUSCO automatically as part of assembly validation [3]. Completeness can be inflated by fragmented matches or reduced by missing genes that are genuinely absent in the sequenced organism.
Contamination Detection
Contamination introduces foreign sequences that inflate assembly size and mislead downstream analyses. Tools like CheckM (for metagenome assembled genomes) or NCBI contamination screening pipelines compare assembly contigs against databases of known contaminants. A small amount of contamination may come from the host microbiome or laboratory reagents. The NCBI Sequence Read Archive holds many raw datasets where contamination was only discovered after assembly [5]. Do not assume a single metric, such as low GC bias, proves the absence of contamination.
Structural Correctness
The ultimate check for structural correctness often requires orthogonal data: genetic maps, HiC proximity ligation, or optical mapping. Chromosome level genome assemblies like those published for Tomicus yunnanensis [8] or Accipiter gularis [10] use HiC to anchor and order scaffolds. Without such validation, even a highly contiguous assembly may contain misjoins or inversions that break gene order. Structural correctness is rarely captured by BUSCO or N50.
Decision Criteria: Which Metric Matters Most for Your Project
Choose your primary quality metric based on the question you are asking.
- If you need a reference for resequencing studies, prioritize structural correctness and completeness. Contiguity is less important than having all genes correctly ordered.
- If you are assembling a novel eukaryotic genome for the first time, aim for high BUSCO completeness and an N50 that exceeds at least one quarter of the estimated chromosome sizes.
- For bacterial genome assembly intended for comparative genomics or typing, prioritize contamination <2% and full BUSCO completeness. N50 above 100 kbp is routine.
- For metagenome assembled genomes (MAGs), contamination and completeness are critical. Use CheckM or similar tools that estimate completeness and contamination from lineage specific marker sets. A MAG with >5% contamination should be treated with caution.
No single metric is sufficient. The Bioconductor project maintains R packages such as gQTL and assemblystats that combine multiple metrics for summary reporting [4].
Practical Workflow for Assembly Quality Assessment
Follow this sequence after you have generated a draft assembly.
- Compute basic statistics. Use command line tools or scripts to get number of contigs, total length, N50, N90, and GC content. Compare total length to expected genome size. NCBI Bookshelf provides detailed formulas for these metrics [1].
- Assess coverage and read mapping. Align a subset of raw reads back to the assembly (e.g., using Minimap2 for long reads or BWA for short reads). Check mapping rate >90% and coverage uniformity. Low mapping may indicate contamination or incomplete assembly.
- Run BUSCO or similar. Execute BUSCO using the appropriate lineage dataset (e.g., eukaryota_odb10, bacteria_odb10). Record complete, fragmented, and missing counts. For bacterial assemblies, you can also use the online NCBI Prokaryotic Genome Annotation Pipeline.
- Check for contamination. Run CheckM (for MAGs) or BLAST contigs against the NCBI nt/nr databases. Flag any contigs with high identity to known contaminants such as human, vector sequences, or common laboratory organisms. Document all flagged contigs.
- Validate structural continuity. If HiC or optical mapping data exist, use tools like 3D DNA (HiC) or Bionano Solve. Compare the assembly to a closely related reference genome using MUMmer or SyRI. Note large scale rearrangements.
- Create a quality report. Combine all metrics into a single table. Include the assembly version, sequencing platform, and software versions. Share this report with the assembly itself when depositing in repositories. The EMBL EBI training resources include templates for such reports [2].
Common Mistakes in Assembly Quality Assessment
- Chasing a high N50 without checking accuracy. Chimeric assemblies are often highly contiguous but completely wrong. Always validate with read pairing or long range information.
- Relying solely on coverage depth. Coverage is easily skewed by duplications, repeats, or poor mapping. Use coverage distribution plots to detect systematic biases.
- Ignoring contaminants that appear as small contigs. Many assemblers discard very short contigs, but contaminants can be long. Screen all contigs above a minimum length (e.g., 500 bp).
- Assuming BUSCO 100% means a perfect assembly. BUSCO markers represent a small fraction of the genome. A complete BUSCO score can hide misassemblies in noncoding regions.
- Not reporting software parameters. Assembly quality is highly dependent on kmer size, read filtering, and error correction choices. Document everything.
Limits and Uncertainty
No assembly is perfect. Even chromosome level assemblies from the Vertebrate Genomes Project contain errors. Limits include:
- Resolution of repetitive regions. Centromeres and telomeres often remain as gaps. The N50 metric ignores these gaps if they fall in the longest contigs.
- Base accuracy in homopolymer runs. Long read assemblers from Oxford Nanopore or PacBio may produce insertion/deletion errors in homopolymers. Coverage alone cannot fix these without a polishing step.
- Polymorphism handling. For diploid or polyploid genomes, assemblers may produce two haplotypes or collapse them. The reported completeness will differ depending on the strategy used.
- Reference bias in completeness assessment. BUSCO relies on curated sets that may not include lineage specific genes. An organism with rapid gene loss or gain will show lower completeness even with a perfect assembly.
When publishing, be transparent about these limitations. Provide the raw QC data and the decision threshold for each metric. The NCBI Sequence Read Archive recommends access to all raw reads so that others can independently verify the assembly [5].
Frequently Asked Questions
1. Should I aim for the highest N50 possible? No. A very high N50 can result from over aggressive merging of unrelated sequences. Instead, aim for an N50 that is consistent with the expected chromosome sizes and validated by long range data. A modest N50 with correct scaffolds is better than a high N50 with misjoins.
2. Can I combine metrics from different assembly tools? Yes, but report each metric as computed by the standard tool. For example, BUSCO version and lineage must be stated because results differ between versions. Do not average or rescale metrics from different programs.
3. What threshold should I use for contamination? For a single isolate bacterial genome, < 1% contamination is typical. For a metagenome assembled genome (MAG), < 5% is often accepted for medium quality genomes, and < 2% for high quality. Always report the contamination detection tool and database.
4. How do I know if my assembly is ready for publication? Check that you have assessed contiguity (N50, L50), completeness (BUSCO), contamination (CheckM or BLAST), and mapping rate. If you claim a chromosome level assembly, you must provide evidence such as HiC contact maps or optical maps. Journals increasingly require that all QC metrics and raw data be deposited.
References and Further Reading
- NCBI Bookshelf: Genome Assembly and Annotation - Authoritative technical references on assembly metrics and quality control [1].
- EMBL EBI Training: Genome Assembly Quality Assessment - Official training modules with practical examples [2].
- Galaxy Training Network: Assembly QC Workflows - Ready to use workflows for BUSCO, Quast, and contamination screening [3].
- Bioconductor: Assembly Quality Packages - R packages for comprehensive assembly evaluation [4].
- NCBI Sequence Read Archive - Repository where raw reads and assemblies are deposited [5].
- ONT only genome assembly of a Korean male using a semen sample - Example of evaluating long read only assembly [6].
- High quality draft genome of Levilactobacillus brevis 3LB - Bacterial assembly QC with CheckM and BUSCO [7].
- Chromosome level genome assembly of Tomicus yunnanensis - Eukaryotic assembly with HiC validation [8].
- Integrated resistome assessment in probiotic products - Example of contamination screening in metagenomic assemblies [11].
Related Articles
- RNA Sequencing Analysis: From FASTQ Files to Biological Questions
- RNA-seq Quality Control: What to Check Before Differential Expression
- How to Plan a Bulk RNA-seq Differential Expression Study
- Single-Cell RNA-seq Workflow: A Practical Analysis Roadmap
- Single-Cell RNA-seq Quality Control: Cells, Genes, and Mitochondrial Reads