Assembly Contamination and Misassembly Detection

  • Assembly contamination and misassembly detection are important stages in genome assembly quality assessment. Genome assembly reconstructs genomic sequences from sequencing reads, but the resulting assembly may contain unwanted DNA, incorrectly joined sequences, duplicated regions, misplaced contigs, or other structural errors. These problems can affect genome annotation, comparative genomics, gene identification, evolutionary analysis, and downstream biological interpretation. Contamination detection aims to identify sequences that originate from unintended organisms or sources, whereas misassembly detection focuses on errors in the reconstruction and organization of the target genome. Both are essential for producing reliable genome assemblies.
  • Assembly contamination occurs when DNA sequences from sources other than the intended target organism are included in the assembled genome. Contaminating sequences may originate from microorganisms associated with the sample, host organisms, laboratory environments, reagents, sequencing adapters, or other biological material. For example, a plant genome assembly may contain bacterial DNA from associated microorganisms, while an animal genome assembly may include sequences from parasites, viruses, or microbial communities. In microbial genome projects, contamination may arise from other bacterial or archaeal species present in the sample, especially when the organism has been isolated from an environmental or clinical source.
  • Contamination can enter a sequencing project at several stages. Biological contamination may occur during sample collection, tissue dissection, DNA extraction, library preparation, or sequencing. Laboratory contamination can result from handling multiple samples in the same environment or using contaminated reagents and equipment. Computational contamination may arise when reads from unintended organisms are included during preprocessing or assembly. Contamination may also occur when sequences from different samples are mistakenly combined during data handling. In metagenomic studies, the presence of multiple organisms is expected, so contamination must be defined relative to the particular genome being reconstructed rather than treating every non-target sequence as unwanted.
  • Contaminating sequences can influence assembly statistics and biological interpretation. They may increase total assembly length, alter GC-content distributions, create additional contigs, and produce misleading gene annotations. Contamination may also affect estimates of genome completeness, gene copy number, metabolic pathways, and evolutionary relationships. In microbial genomics, contaminating sequences can create the false impression that an organism contains genes or pathways that actually originate from another species. In clinical and environmental studies, contamination can complicate the identification of pathogens, antimicrobial resistance genes, and other biologically important features.
  • Misassembly occurs when genomic sequences are incorrectly reconstructed or organized during genome assembly. It may involve incorrect joins between unrelated regions, inversions, translocations, duplicated segments, collapsed repeats, missing regions, or incorrect sequence orientation. A misassembled contig may contain sequence from different genomic locations joined together, while a scaffold may incorrectly connect contigs or place them in the wrong order. Misassemblies can occur because sequencing reads do not provide enough information to distinguish repetitive or similar genomic regions, because sequencing errors interfere with overlap detection, or because the assembly algorithm makes an incorrect choice among alternative paths in an assembly graph.
  • One common type of misassembly is an incorrect contig join. This occurs when two genomic regions that are not adjacent in the target genome are joined into a single continuous sequence. Such errors may arise when repeated sequences cause reads from different genomic locations to be incorrectly connected. Incorrect joins can also result from contamination, insufficient read coverage, or ambiguous assembly graph structures. Long-read sequencing and paired-end information can help resolve these ambiguities, but they do not eliminate all possibilities of incorrect reconstruction.
  • Inversions and translocations are other important forms of structural misassembly. An inversion occurs when a genomic region is assembled in the reverse orientation relative to its correct position. A translocation occurs when a region is incorrectly placed or joined with another genomic location. These errors may be difficult to detect using short reads alone because short sequencing fragments often provide limited information about long-range genomic organization. Long reads, Hi-C data, optical maps, and genetic maps can provide additional evidence for detecting these structural problems.
  • Repeat-associated misassemblies are particularly common in complex genomes. Repetitive DNA sequences occur in multiple locations and may be highly similar or identical. When reads do not extend beyond the repeated region, the assembler may be unable to determine which copy belongs at a particular location. This can lead to repeat collapse, in which multiple repeat copies are represented as a single sequence, or repeat expansion, in which the same region is represented more than once. Repeats may also cause incorrect connections between genomic regions. These problems are especially relevant in plant and animal genomes, which often contain large amounts of repetitive DNA.
  • Haplotype-related errors can also complicate assembly evaluation. In diploid and polyploid organisms, homologous chromosomes may contain similar but non-identical sequences. Some assembly approaches collapse homologous regions into a single representation, while others retain separate haplotypes. Inconsistent representation can produce duplicated regions, artificial joins, or apparent missing sequences. Researchers should distinguish genuine biological differences between haplotypes from technical duplication or misassembly. The intended representation of the genome should be defined before interpreting duplicated sequence regions.
  • Contamination detection often begins with sequence composition analysis. GC content represents the proportion of guanine and cytosine nucleotides in a DNA sequence. Different organisms and genomic regions may have characteristic GC-content distributions. If a subset of contigs has a GC content substantially different from the main assembly, those contigs may warrant further investigation. However, GC content alone is not sufficient to identify contamination because genuine genomes can contain regions with unusual nucleotide composition, including horizontally transferred genes, organellar DNA, and regions with different evolutionary histories.
  • Sequencing coverage provides another useful signal. Coverage describes how many sequencing reads, on average, support each position in the assembled genome. Contigs from contaminating organisms may have different coverage patterns from the target genome. For example, a contaminating organism present at low abundance may produce contigs with lower coverage, while a highly abundant contaminant may produce contigs with unusually high coverage. Coverage differences can also result from biological copy-number variation, repetitive sequences, GC bias, uneven sequencing, or differences between chromosomes and plasmids. Coverage should therefore be interpreted alongside other evidence.
  • Taxonomic classification is one of the most effective approaches for investigating contamination. Assembled contigs or sequencing reads can be compared against reference databases to identify their likely taxonomic origins. Tools such as Kraken2, Centrifuge, Kaiju, and BLAST can help classify sequences based on nucleotide or protein similarity. Kraken2 uses k-mer-based classification, while Kaiju uses protein-level sequence information. BLAST can provide more detailed similarity searches for selected sequences. Classification results depend on the quality and completeness of the reference databases, and highly divergent or poorly represented organisms may be assigned incorrectly or remain unclassified.
  • The NCBI nucleotide and protein databases, RefSeq, UniProt, and other curated resources can support taxonomic investigation. Researchers should consider whether the reference database contains appropriate representatives of the target organism and likely contaminants. A sequence classified as bacterial, for example, may represent a genuine bacterial symbiont, horizontally transferred DNA, an organellar sequence, or contamination. Taxonomic classification is therefore evidence for investigation rather than automatic proof that a sequence should be removed.
  • BlobToolKit is a widely used resource for investigating genome assembly composition and possible contamination. It combines information such as GC content, sequencing coverage, sequence length, and taxonomic assignments to visualize contigs in a multidimensional context. Contigs with unusual combinations of these properties can be examined as potential contaminants. BlobToolKit is particularly useful for large and complex assemblies because it allows researchers to inspect groups of sequences rather than relying on a single threshold. The biological origin of suspicious sequences should still be verified before removal.
  • Other tools can support contamination screening. CheckM and CheckM2 are commonly used for estimating completeness and contamination in bacterial and archaeal genomes, particularly metagenome-assembled genomes. They use marker-gene information and computational models to evaluate whether a genome contains expected genes and whether some markers occur in patterns suggesting contamination or duplication. BUSCO can also identify duplicated conserved genes, although duplicated BUSCOs are not direct proof of contamination. Genuine gene duplication, polyploidy, and haplotype-resolved assemblies can produce similar patterns.
  • Misassembly detection relies on several complementary types of evidence. Read alignment is a common first step. Sequencing reads are mapped back to the assembled genome, and the alignment patterns are examined for inconsistencies. A sudden drop in read coverage may indicate a missing or incorrectly assembled region. Abrupt changes in coverage, clusters of mismatches, unusual insertions or deletions, and inconsistent read-pair orientations may suggest a structural problem. However, these signals can also arise from repetitive DNA, biological variation, sequencing errors, or mapping limitations.
  • Paired-end sequencing data provide useful information about the expected distance and orientation between reads originating from the same DNA fragment. If read pairs map to an assembly in unexpected orientations or at unusually large or small distances, the region may contain a misassembly. For example, pairs that should map with a consistent inward-facing orientation may instead map in the same orientation or on different scaffolds. Clusters of such discordant read pairs can indicate incorrect joins, inversions, or rearrangements. These signals are most informative when the library insert-size distribution and sequencing characteristics are known.
  • Long-read sequencing provides additional evidence for structural validation. A correctly assembled region should generally be supported by reads that span the relevant sequence or connect adjacent regions consistently. If long reads repeatedly align across an apparent join with substantial discrepancies, or if the assembly contains a junction unsupported by the raw reads, the region may be misassembled. Long-read alignments can also help identify collapsed repeats, incorrect repeat copy numbers, and unresolved haplotypes. The interpretation depends on read length, accuracy, coverage, and the complexity of the region.
  • Hi-C sequencing provides long-range information about physical interactions between genomic regions. In chromosome-scale assembly, Hi-C contact patterns can help evaluate whether scaffolds have been ordered and oriented correctly. Regions belonging to the same chromosome often show characteristic interaction patterns, while incorrectly joined or misplaced sequences may produce unexpected contact patterns. Tools such as Juicer, 3D-DNA, and YaHS-related workflows can be used in Hi-C-based scaffolding and evaluation. Hi-C evidence is particularly useful for detecting large-scale structural problems, but it should be interpreted with the limitations of chromatin interaction data in mind.
  • Optical mapping and genetic maps can also help validate large-scale genome structure. Optical mapping provides information about the relative positions of restriction or labelling patterns along long DNA molecules. Genetic maps use recombination information to estimate the relative positions of markers. If an assembled sequence conflicts with these independent maps, the discrepancy may indicate a structural misassembly or a genuine biological rearrangement. These methods are especially valuable for large plant and animal genomes where chromosome organization is important.
  • Reference-based comparison is another important method for identifying misassemblies. When a closely related, high-quality reference genome is available, the assembled sequences can be aligned against it. Large alignment gaps, unexpected changes in orientation, inconsistent sequence order, and abrupt shifts in genomic position may indicate structural differences or assembly errors. Tools such as QUAST and MUMmer can support these comparisons. Minimap2 is commonly used for aligning long sequences and reads. Reference-based detection must be interpreted carefully because real evolutionary rearrangements and sample-specific structural variants may differ from the reference without representing assembly mistakes.
  • QUAST (Quality Assessment Tool for Genome Assemblies) is widely used to evaluate genome assemblies and identify potential misassemblies when a suitable reference is available. It reports continuity statistics and performs reference-based comparisons that can reveal inconsistencies between the assembly and reference. The following command illustrates a basic assessment:

quast.py assembly.fasta -o quast_results

  • If a suitable reference genome is available, it can be included:

quast.py assembly.fasta \

  -r reference.fasta \

  -o quast_reference_results

  • The reference-based report may identify regions where the assembly differs substantially from the reference in sequence order, orientation, or genomic position. Such regions should be investigated rather than automatically classified as errors, because genuine biological rearrangements may explain some differences.
  • Read mapping is a useful way to investigate suspicious assembly regions. The following example uses minimap2 and SAMtools to map long reads against an assembly and generate an indexed BAM file:

minimap2 -ax map-ont assembly.fasta reads.fastq.gz | \

  samtools sort -o reads_mapped.bam

samtools index reads_mapped.bam

  • The map-ont preset is intended for Oxford Nanopore reads. PacBio HiFi reads generally require a different preset, such as map-hifi. The alignment can then be inspected using visualization software such as IGV or other genome browsers. Coverage and alignment patterns may reveal regions that require further investigation. The mapping parameters should be adapted to the sequencing technology and data characteristics.
  • Coverage can be summarized using SAMtools:

samtools coverage reads_mapped.bam

  • This command provides coverage-related statistics for reference sequences. For detailed regional analysis, depth information can also be generated:

samtools depth -a reads_mapped.bam > coverage.txt

  • Regions with unusually low or high coverage can then be examined alongside read alignment patterns, GC content, and taxonomic assignments. Coverage differences alone should not be treated as proof of contamination or misassembly.
  • Kraken2 can be used to classify sequences taxonomically and identify potential contaminating organisms. A simplified example for classifying sequencing reads is:

kraken2 \

  –db kraken2_database \

  –threads 8 \

  –report kraken_report.txt \

  –output kraken_output.txt \

  reads.fastq.gz

For assembled contigs, Kraken2 can classify the sequences directly:

kraken2 \

  –db kraken2_database \

  –threads 8 \

  –report contig_report.txt \

  –output contig_classification.txt \

  –fasta-input \

  assembly.fasta

  • The database should contain suitable reference sequences for the organisms under investigation. Classification accuracy depends on database composition, sequence similarity, and the taxonomic diversity represented. A contig assigned to an unexpected taxon should be investigated using additional evidence before it is removed.
  • BlobToolKit can help combine sequence composition, coverage, and taxonomic information for visual inspection. A typical workflow involves preparing an assembly dataset, obtaining coverage and taxonomic annotations, and exploring the resulting data through interactive plots. The precise commands and input formats depend on the installed BlobToolKit version and the workflow used to generate the dataset. Its visual summaries are particularly useful for identifying groups of contigs with unusual properties that may represent contaminants or unusual regions of the target genome.
  • CheckM2 can be used to assess completeness and contamination in microbial genome assemblies. A simplified example is:

checkm2 predict \

  –input assembly_directory \

  –output-directory checkm2_results \

  –threads 8

  • The input directory and file organization should follow the requirements of the installed CheckM2 version. CheckM2 is primarily intended for microbial genomes and should not be treated as a general contamination detector for all eukaryotic assemblies. Its estimates should be interpreted alongside taxonomic classification, assembly statistics, and biological information.
  • When contamination or misassembly is detected, corrective action should be based on the evidence supporting the suspected problem. Contaminating contigs may be removed when their non-target origin is established through multiple independent observations. However, suspicious sequences should not be discarded solely because they have unusual GC content, coverage, or taxonomic assignments. Genuine organellar sequences, symbionts, horizontally transferred genes, and divergent genomic regions may share these characteristics.
  • Misassembled regions may require local correction, reassembly, or scaffolding adjustments. If an incorrect contig join is supported by clear evidence, the sequence may be broken at the suspected junction and reconstructed using the original reads. Repeat-associated errors may require longer reads or additional long-range data. Scaffolding errors may be corrected by reassessing the order and orientation of contigs using Hi-C, optical mapping, genetic maps, or other independent evidence. In difficult regions, retaining unresolved gaps may be more reliable than introducing unsupported sequence.
  • Assembly graphs can provide useful information during misassembly investigation. An assembly graph represents alternative connections between sequence regions and can reveal unresolved repeats, branches, bubbles, and possible incorrect joins. Tools such as Bandage allow researchers to visualize assembly graphs in formats such as GFA. A suspicious region in the final linear assembly can be examined in the graph to determine whether alternative paths or ambiguous connections were present during assembly. Graph inspection should be combined with read evidence because graph complexity alone does not establish that an assembly is incorrect.
  • Contamination and misassembly detection are particularly important in metagenomic studies. Metagenome-assembled genomes may contain sequences from multiple closely related organisms, especially when the community includes strains with similar genomes. Binning tools such as MetaBAT2, MaxBin2, CONCOCT, and VAMB group contigs into genome bins using features such as coverage and sequence composition. However, bins may contain contaminating contigs or may combine sequences from related organisms. Completeness and contamination estimates, taxonomic classification, coverage across multiple samples, and marker-gene analysis are therefore important for evaluating metagenome-assembled genomes.
  • Several limitations affect contamination and misassembly detection. Reference databases may not contain close relatives of the target organism or contaminating species. Short reads may not span repetitive regions or structural junctions. Long reads may contain sequencing errors, and uneven coverage may complicate alignment interpretation. Genuine biological variation may resemble an assembly error when compared with a reference genome. In addition, contamination from closely related organisms may be difficult to distinguish using sequence similarity alone. These limitations make it important to combine independent evidence and avoid relying on a single quality metric or detection tool.
  • A comprehensive genome assembly validation workflow should begin with raw sequencing data quality assessment and contamination screening. After assembly, researchers should evaluate continuity using N50, L50, and related statistics; completeness using BUSCO or other suitable methods; and contamination using taxonomic classification, sequence composition, and microbial genome quality tools where appropriate. Read mapping, reference comparisons, k-mer analysis, and long-range evidence should then be used to investigate potential misassemblies. Suspicious regions should be reviewed, corrected when justified, and reassessed after any changes to the assembly.
  • Reproducibility is essential throughout this process. Researchers should document the sequencing technologies, read types, assembly software and version, parameters, reference databases, database versions, and quality assessment tools used. The criteria applied to flag or remove suspicious contigs should also be recorded. Original assembly files should be preserved so that corrections can be traced and independently reviewed. Final reports should distinguish between confirmed contamination, suspected contamination, reference differences, and potential structural errors rather than presenting all unusual regions as definite assembly failures.
  • Assembly contamination and misassembly detection are fundamental to producing reliable genome assemblies. Contamination screening identifies sequences that may originate from unintended organisms or sources, while misassembly detection identifies errors in sequence reconstruction and genomic organization. Tools such as Kraken2, BlobToolKit, CheckM2, QUAST, minimap2, and SAMtools provide complementary evidence for investigating these problems. Their results should be interpreted alongside sequencing reads, assembly graphs, reference comparisons, and biological knowledge. A high-quality genome assembly requires not only good continuity and completeness but also reliable sequence origin, nucleotide accuracy, and correct structural organization.
Author:

Leave a Reply

Your email address will not be published. Required fields are marked *