- Repeat resolution and assembly graph interpretation are important aspects of genome assembly that help reconstruct complex genomic regions and improve the accuracy of assembled sequences. Many genomes contain repetitive DNA sequences, including tandem repeats, interspersed repeats, transposable elements, segmental duplications, and duplicated genes. These sequences create difficulties for genome assembly because different genomic regions may contain identical or highly similar sequences. When sequencing reads cannot distinguish between repeat copies, assembly algorithms may fail to determine their correct genomic positions and connections. Assembly graphs provide a representation of these alternative sequence relationships, while repeat-resolution methods use additional sequencing evidence and computational approaches to determine how repeated regions should be reconstructed.
- Repetitive DNA occurs throughout the genomes of bacteria, plants, animals, and other organisms. Some repeats are short and occur many times, while others extend over thousands or millions of nucleotide positions. Tandem repeats consist of sequences arranged consecutively, such as microsatellites and satellite DNA. Interspersed repeats occur at different genomic locations and include transposable elements and their remnants. Segmental duplications are relatively large DNA regions that have been copied within a genome and may retain high sequence similarity. Gene families also contain duplicated sequences that can complicate assembly, particularly when multiple copies are nearly identical.
- Repeats create a fundamental challenge because genome assembly depends on identifying overlaps or shared sequence information among reads. If a repeat is longer than the sequencing reads or the reliable region of overlap between them, the assembler may be unable to determine which repeat copy connects to which unique genomic region. For example, suppose a genome contains two unique regions separated by an identical repeat sequence at different locations. Reads that cover only the repeat cannot reveal which copy they originated from. The assembler may therefore collapse both copies into a single sequence, leave the region unresolved, or connect the surrounding sequences incorrectly.
- Repeat length is an important factor in determining whether a repeated region can be resolved. Short-read technologies such as Illumina sequencing generate highly accurate reads, but their relatively short length may prevent them from spanning long repetitive regions. Long-read technologies, including PacBio HiFi and Oxford Nanopore sequencing, can produce reads that extend across longer repeats and into unique flanking regions. When a read spans an entire repeat and includes sufficient unique sequence on both sides, it can provide direct evidence for the correct genomic connection. However, long reads do not automatically resolve every repeat. Very long repeats, low-complexity regions, insufficient coverage, sequencing errors, and highly similar duplicated regions can still create ambiguity.
- Repeat resolution aims to determine the correct number, location, orientation, and connectivity of repeated genomic sequences. It is closely associated with the detection of repeat copies, identification of unique flanking regions, analysis of read coverage, and reconstruction of the genomic paths through repeated regions. An assembler may initially represent a repeat as a shared sequence segment connected to multiple possible genomic paths. Additional evidence is then required to determine which paths correspond to the actual genome.
- Assembly graphs are computational structures used to represent the relationships among sequences during genome assembly. Instead of immediately producing a single linear genome sequence, an assembler can represent possible sequence connections as a graph. In a graph, nodes or edges represent sequence information, while connections represent possible overlaps, extensions, or transitions between sequences. Different assembly algorithms use different graph structures, so the meaning of nodes, edges, directions, and labels depends on the underlying representation.
- Two major graph approaches used in genome assembly are the overlap-layout-consensus approach and de Bruijn graphs. In overlap-layout-consensus assembly, reads are represented as graph elements, and edges indicate significant overlaps between reads. The layout phase determines how overlapping reads can be arranged into longer sequences, while the consensus phase estimates the most likely nucleotide sequence from the aligned reads. Overlap-based graphs are particularly important in long-read assembly, where individual reads may cover substantial portions of the genome.
- De Bruijn graphs are widely used in short-read genome assembly. In this approach, sequencing reads are divided into smaller overlapping sequences called k-mers. Each k-mer has a length of k nucleotides. Depending on the implementation, graph nodes may represent k-mers of length k−1, while edges represent k-mers that connect adjacent nodes. A path through the graph corresponds to a possible assembled sequence. The graph can therefore represent multiple possible arrangements of the original reads without requiring the assembler to choose a single sequence immediately.
- For example, the DNA sequence ATGCTA can be divided into 3-mers as follows:
- ATG
- TGC
- GCT
- CTA
- In a simplified de Bruijn graph representation using 2-mer nodes, these k-mers connect the following sequence fragments:
- AT -> TG
- TG -> GC
- GC -> CT
- CT -> TA
- The resulting path reconstructs the original sequence. In real genome assembly, the graph contains many overlapping paths generated from millions of k-mers, and repeated sequences may create branches or shared graph structures.
- Assembly graphs are especially useful for understanding repetitive regions because they preserve alternative connections that may be lost when an assembler outputs only a linear sequence. A repeated sequence may appear as a shared graph segment connected to several possible paths. If the repeat has multiple copies in the genome, the graph may contain branching structures indicating alternative routes through the assembly. These branches can represent genuine genomic alternatives, unresolved repeats, sequencing errors, heterozygous variants, structural differences, or combinations of these features.
- A bubble is a common structure in an assembly graph. It occurs when two alternative paths connect the same starting and ending regions. Bubbles may arise from sequencing errors, heterozygous variants, small insertions or deletions, or alternative sequence arrangements. In a diploid genome, a bubble may represent differences between maternal and paternal haplotypes. However, not every bubble represents a true biological variant. Some may be caused by errors or insufficiently resolved repeats. Interpreting a bubble therefore requires examination of read support, sequence similarity, coverage, and the biological context.
- Branches occur when a graph node or edge has multiple possible connections. A branching point may indicate that the sequence can continue along more than one path. In repeat-rich regions, branches commonly arise because repeated sequences connect to several different genomic locations. The assembler may not have enough information to determine the correct path. A branch can also arise from sequencing errors, polymorphisms, or structural variation. The presence of a branch does not necessarily indicate an assembly error; it may represent genuine genomic complexity.
- Dead ends are graph paths that terminate without an obvious continuation. They may result from low sequencing coverage, sequencing errors, incomplete reads, contamination, or genuine chromosome ends. In some cases, a dead end indicates that the assembler cannot identify sufficient overlapping sequence to continue the assembly. A high number of dead ends may indicate poor read quality, inadequate coverage, or unresolved genomic regions. However, their interpretation depends on the graph construction method and the organism’s genome structure.
- Coverage is an important feature for interpreting assembly graphs. Coverage describes how many sequencing reads or k-mer observations support a particular sequence region. Repeated sequences may show higher coverage than unique regions because multiple genomic copies contribute reads with similar or identical sequences. For example, if a repeat occurs five times in a genome, reads from all five copies may map to a single collapsed repeat representation. This can produce unusually high coverage compared with neighbouring unique regions. However, coverage differences can also result from uneven sequencing depth, GC bias, amplification bias, contamination, or differences in copy number. Coverage alone is therefore insufficient to determine the exact number of repeat copies.
- Repeat resolution methods use several types of evidence to distinguish repeated regions. Read length is particularly important because reads spanning a repeat and its unique flanking sequences can identify the correct connection. Paired-end and mate-pair reads can provide linking information when their fragment lengths are sufficient to connect regions on either side of a repeat. Long reads can provide direct sequence continuity across repeats. Hi-C data can support long-range chromosome organization, although physical proximity information does not always establish direct sequence adjacency. Optical maps and genetic linkage information can provide additional evidence about large-scale genomic structure.
- Long-read sequencing has substantially improved the resolution of repetitive genomic regions. PacBio HiFi sequencing combines long reads with high consensus accuracy, making it useful for distinguishing similar repeat copies and reconstructing complex genomic regions. Oxford Nanopore sequencing can produce very long reads that span large repeats and structural variations. The effectiveness of either technology depends on the length and accuracy of the reads, the coverage of the region, and the uniqueness of the sequences surrounding the repeat. Hybrid assembly approaches combine long-read structural information with short-read accuracy, although the integration strategy must be appropriate for the technologies and organism being analysed.
- Repeat resolution also depends on the ability to distinguish genuine repeat copies from collapsed or duplicated assembly sequences. Repeat collapse occurs when multiple genomic copies are represented as a single assembled sequence. This can lead to underestimated genome size, incorrect gene copy numbers, and distorted coverage. Haplotig duplication occurs when alternative haplotypes are assembled as separate sequences even though they represent corresponding regions of homologous chromosomes. This can inflate the apparent assembly size and create redundant sequences. These problems are especially important in diploid and polyploid genomes, where multiple homologous or related chromosome copies must be distinguished.
- Graph simplification is a computational process used to reduce unnecessary complexity in assembly graphs. It may involve removing low-support paths, correcting sequencing-error-derived branches, merging unambiguous linear paths, and resolving small bubbles. The goal is to simplify the graph while retaining genuine biological variation and alternative genomic structures. Excessive simplification can remove real variants or structural differences, while insufficient simplification may leave too many ambiguous paths for reliable assembly. The appropriate approach depends on sequencing accuracy, coverage, genome complexity, and the purpose of the analysis.
- Several bioinformatics tools provide assembly graph generation, visualization, and analysis. SPAdes generates assembly graphs for short-read and some hybrid assembly workflows and provides graph files that can be inspected using Bandage. MEGAHIT is widely used for metagenomic assembly and generates graph-based assembly information. Velvet is an earlier de Bruijn graph assembler for short reads. Flye constructs repeat graphs for long-read assembly, while Canu uses overlap-based methods for long-read genome assembly. Hifiasm is widely used for high-quality PacBio HiFi assembly and can represent alternative haplotype paths. Verkko combines long-read and other evidence to assemble complex genomes, including haplotype-aware assemblies.
- Bandage is a widely used graphical application for visualizing assembly graphs. It can display graph nodes, connections, sequence lengths, coverage information, and other available attributes. Researchers can use it to inspect branching regions, circular structures, disconnected components, and possible repeat-associated ambiguities. Bandage supports graph formats generated by several assembly tools, including GFA-based representations. Bandage and related graph viewers help researchers understand the structure of an assembly, but visual inspection alone cannot prove that a particular path is biologically correct.
- The GFA format, or Graphical Fragment Assembly format, is a text-based format used to represent sequence graphs. It can describe sequence segments, links between segments, paths through the graph, and other graph-related information. GFA is commonly used to exchange assembly graph information between compatible tools. The exact record types and supported features depend on the GFA version. A simplified illustrative example is:
H VN:Z:1.0
S 1 ATGCTA
S 2 GGATCC
L 1 + 2 + 0M
P path1 1+,2+ *
- In this example, H identifies the header, S records describe sequence segments, L describes a link between two segments and their orientations, and P describes a path through the graph. The example is intended to illustrate the basic structure rather than represent a complete genome assembly graph. Real GFA files may contain many segments, links, paths, tags, and other information.
- Graph interpretation often requires identifying whether a path represents a unique genomic sequence, an alternative haplotype, a repeat copy, or an unresolved connection. Researchers should examine node and edge coverage, sequence length, read support, orientation, and the consistency of long-range links. A path supported by many independent reads and consistent with the expected genome structure is generally more reliable than a path supported by a small number of ambiguous alignments. However, the number of supporting reads must be interpreted in relation to sequencing depth, library design, and the possibility of mapping bias.
- The following command illustrates how an assembly graph generated by SPAdes can be visualized using Bandage:
Bandage image assembly_graph.gfa assembly_graph.png
- The exact graph filename depends on the assembler and its output configuration. SPAdes commonly produces an assembly graph in GFA-compatible or FASTG form, depending on the output file. Bandage supports multiple graph formats, but users should verify compatibility with the specific file they intend to inspect. The resulting image can provide an overview of graph connectivity and help identify complex regions that may require further analysis.
- For long-read assembly, Flye produces assembly graph information that can be inspected using compatible graph visualization tools. A simplified Flye command for Oxford Nanopore reads is:
flye –nano-raw long_reads.fastq.gz \
–out-dir flye_output \
–threads 16 \
–genome-size 5m
- The –genome-size value is an approximate estimate of the genome size and should be adjusted to the organism being assembled. Other read types, including PacBio HiFi, require the appropriate Flye input mode. The assembly graph and associated outputs can then be examined to identify unresolved repeats, alternative paths, and possible structural ambiguities.
- Repeat resolution is particularly important in metagenomic assembly, where sequencing data originate from multiple organisms with different abundances and genome structures. Closely related microbial strains may share large portions of their genomes while differing in particular regions. Repeated genes, mobile genetic elements, plasmids, and conserved genomic regions can create ambiguous graph connections. In such cases, an assembly graph may contain paths representing different organisms or strains. Incorrectly resolving these paths can produce chimeric contigs or incorrectly combine sequences from different genomes. Coverage patterns, sequence composition, taxonomic information, long reads, and multi-sample evidence can help distinguish these alternatives.
- In eukaryotic genome assembly, repeat resolution is important for reconstructing centromeres, telomeres, segmental duplications, transposable-element-rich regions, and duplicated gene families. These regions may be difficult to assemble because they contain long sequences with high similarity. Long-read sequencing, Hi-C, optical mapping, and specialized assembly algorithms have improved the reconstruction of such regions. Nevertheless, some highly repetitive regions remain unresolved even in advanced genome assemblies. The presence of gaps or alternative graph paths should therefore be documented rather than hidden by unsupported sequence reconstruction.
- Assembly graph interpretation is also important in pangenome analysis. A pangenome graph represents sequence variation and alternative genomic paths across multiple individuals, strains, or populations. Unlike a conventional assembly graph that focuses on reconstructing one genome, a pangenome graph can represent shared and variable genomic regions across many genomes. Such graphs are useful for studying structural variation, gene presence and absence, population diversity, and evolutionary relationships. However, pangenome graphs and single-genome assembly graphs have different purposes and should not be treated as interchangeable.
- The quality of repeat resolution and graph interpretation can be assessed using several complementary approaches. Read mapping can determine whether sequencing reads support the proposed paths and connections. Coverage analysis can identify possible repeat collapse or unexpected copy-number differences. BUSCO can assess the completeness of conserved gene content, although it does not directly measure repeat resolution or structural correctness. Merqury can evaluate assembly quality using k-mer-based methods, including completeness and consensus accuracy. QUAST can provide assembly statistics and identify some potential inconsistencies when an appropriate reference is available. For chromosome-scale assemblies, Hi-C contact maps and optical maps can provide additional evidence for large-scale organization.
- Researchers should also compare the assembly graph with the final linear assembly. A linear FASTA file may contain only one selected path through a graph, while the graph preserves alternative connections that were not included in the final output. Examining both representations can reveal unresolved branches, collapsed repeats, alternative haplotypes, and possible misassemblies. When a graph contains several plausible paths, the final assembly should not be interpreted as uniquely determined unless the available evidence supports that conclusion.
- Repeat resolution and assembly graph interpretation have important limitations. Short reads may not span long repeats, long reads may contain errors or fail to cover particular regions, and sequencing coverage may be uneven. Highly similar duplicated regions may remain indistinguishable even with long reads. In diploid genomes, alternative haplotypes can produce graph complexity that resembles repeat-related branching. In metagenomes, closely related species and strains may create ambiguous paths. Graph visualization also depends on the graph construction method and software conventions, so a particular visual structure may have different meanings in different assembly systems.
- For reproducible genome assembly, researchers should preserve the original assembly graph, final contig and scaffold sequences, raw sequencing data, software names and versions, command-line parameters, and relevant intermediate files. They should document how repeat-associated ambiguities were evaluated, which paths were selected, and what evidence supported major connections. Unresolved regions should be reported transparently, particularly when they affect gene copy number, chromosome structure, structural variation, or biological interpretation. When possible, independent technologies should be used to validate important repeat resolutions and large-scale assembly structures.
- Repeat resolution and assembly graph interpretation are therefore essential for reconstructing accurate and complete genomes. Repetitive DNA creates alternative sequence connections that cannot always be resolved from short sequencing reads alone. Assembly graphs preserve these alternatives and provide a framework for examining sequence connectivity, while long reads, paired-end information, Hi-C data, optical maps, and other evidence help determine the most reliable genomic paths. Understanding graph structures, repeat-associated ambiguities, coverage patterns, and validation methods enables researchers to evaluate genome assemblies more critically and produce more reliable genomic resources for evolutionary biology, comparative genomics, medicine, agriculture, and microbial research.