- Reference bias and assembly limitations are important considerations in genome analysis because they influence how accurately biological sequence information is reconstructed and interpreted. Genome assembly attempts to reconstruct genomic sequences from sequencing reads, either by using a reference genome or by assembling reads without relying on an existing reference. Although modern sequencing technologies and assembly algorithms have greatly improved genome reconstruction, several technical and biological challenges remain. Reference bias occurs when computational methods preferentially represent sequences that resemble an available reference genome, potentially underrepresenting genuine differences in the sample. Assembly limitations arise from the characteristics of sequencing data, genome complexity, computational algorithms, and biological variation that make complete and accurate reconstruction difficult.
- Reference bias is particularly important in reference-guided genome assembly and reference-based sequence analysis. In reference-guided assembly, sequencing reads are aligned to an existing reference genome, and the resulting alignment information is used to reconstruct or refine the target sequence. Reads that closely resemble the reference generally map more easily than reads containing substantial sequence differences. Consequently, highly divergent regions, insertions, deletions, structural variants, and sequences absent from the reference may be poorly represented or excluded from the reconstructed genome. The resulting assembly can therefore resemble the reference more closely than the actual genome of the sample.
- Reference bias can occur during read alignment. Most alignment algorithms use scoring systems that evaluate matches, mismatches, insertions, and deletions. Reads containing many differences from the reference may receive lower alignment scores, map ambiguously, or remain unmapped. If the analysis retains only confidently mapped reads, genuine sample-specific sequences may be lost. This problem is particularly relevant when analysing genetically diverse populations, highly polymorphic genomic regions, or organisms that are evolutionarily distant from the reference.
- Reference bias can also affect variant detection. When sequencing reads are mapped to a reference genome, variants are identified by comparing the sample’s reads with the reference sequence. Reads carrying alleles that differ substantially from the reference may map less efficiently than reads carrying reference-like alleles. This can lead to underrepresentation of alternative alleles, incorrect genotype calls, and reduced sensitivity for detecting certain variants. In some circumstances, the reference allele may appear more frequent than it actually is because reads carrying the alternative allele are less likely to align successfully.
- Allelic mapping bias is a related problem in which reads carrying one allele are preferentially mapped over reads carrying another allele. This can distort estimates of allele-specific expression, allele frequency, and genetic variation. It is especially important in studies of gene expression, population genetics, cancer genomics, and comparative genomics. For example, if reads from one parental chromosome contain more sequence differences from the reference than reads from the other chromosome, the alignment process may favour the reference-like chromosome. The resulting data may incorrectly suggest unequal expression or allele representation.
- Reference bias can also affect the analysis of structural variation. Large insertions, deletions, inversions, duplications, and translocations may not be adequately represented by a linear reference genome. Reads containing sequences absent from the reference may fail to align or may align only partially. Complex structural variants can also produce ambiguous mappings because multiple genomic regions share similar sequences. Consequently, reference-based analyses may underestimate structural diversity, particularly in repetitive regions and genomic regions that differ substantially among individuals or populations.
- The choice of reference genome is therefore important. A reference should be sufficiently close to the organism or population under investigation and should have high sequence quality, completeness, and annotation reliability. However, even a high-quality reference genome represents a particular biological sequence rather than every possible genomic variation within a species. Human reference assemblies, for example, cannot represent all genetic differences among human populations. Similarly, a reference genome from one plant or animal breed may not adequately represent the genomic diversity of other breeds or populations. Researchers should therefore avoid treating a single reference genome as a universal representation of a species.
- Reference-guided assembly has several advantages despite these limitations. It can reduce computational complexity, support the reconstruction of genomes from relatively limited sequencing data, and provide a useful coordinate framework for comparative analysis. It is particularly effective when a closely related, high-quality reference genome is available and the target genome is not expected to differ substantially. However, its performance decreases when the target genome contains substantial divergence, large structural changes, novel sequences, or complex repetitive regions.
- De novo genome assembly provides an alternative approach because it reconstructs genomic sequences directly from sequencing reads without requiring an existing reference genome. This can reduce dependence on a particular reference and allow the discovery of novel sequences, structural differences, and genomic regions absent from established references. De novo assembly is particularly useful for newly sequenced species, highly diverse populations, organisms with substantial genomic rearrangements, and studies that aim to develop new reference genomes. Nevertheless, de novo assembly has its own limitations, including high computational requirements, incomplete reconstruction, repeat-resolution problems, and difficulties with heterozygous or polyploid genomes.
- One major limitation of genome assembly is repetitive DNA. Repetitive sequences occur in multiple copies throughout the genome and may be identical or highly similar. These include tandem repeats, transposable elements, segmental duplications, satellite DNA, and repeated gene families. If sequencing reads are shorter than the repeated regions, they may not contain enough unique information to determine which genomic copy they originate from. Assembly algorithms may therefore collapse multiple repeat copies into a single sequence, duplicate repeated regions, or leave them unresolved. Repetitive regions are a major source of fragmentation and structural uncertainty, particularly in large plant and animal genomes.
- Genome heterozygosity is another important assembly challenge. Heterozygosity refers to the presence of different alleles at corresponding genomic positions on homologous chromosomes. In diploid organisms, sequencing reads may represent both parental chromosome copies, which can differ across many regions. Some assemblers may incorrectly treat these allelic differences as separate genomic sequences, producing redundant contigs known as haplotigs. Other assemblers may collapse the two haplotypes into a single consensus sequence, potentially losing genuine allele-specific information. The appropriate strategy depends on whether the research objective requires a collapsed representation or separate haplotype-resolved assemblies.
- Polyploid genomes introduce additional complexity because they contain more than two chromosome sets. In plants and some other organisms, multiple related chromosome copies may contain similar but non-identical sequences. Assembly algorithms may struggle to distinguish homologous and homeologous regions, leading to collapsed sequences, duplicated contigs, or incorrect connections. Accurate reconstruction may require specialized polyploid-aware or haplotype-aware assembly strategies and additional long-range evidence.
- Sequencing read length and accuracy also influence assembly quality. Short-read sequencing technologies generally provide highly accurate reads but may not span long repeats or complex structural regions. Long-read technologies can span larger genomic regions and improve repeat resolution, but their error profiles and coverage characteristics vary by platform and sequencing method. PacBio HiFi reads combine long-read sequencing with high consensus accuracy, while Oxford Nanopore technologies can produce very long reads that help connect distant genomic regions. Nevertheless, long reads may still be insufficient to resolve exceptionally long repeats, highly similar duplicated regions, or complex chromosome structures.
- Sequencing coverage is another important limitation. Coverage describes how many sequencing reads, on average, represent each genomic position. Insufficient coverage may prevent the assembly algorithm from identifying reliable overlaps or reconstructing low-abundance regions. Uneven coverage can also create fragmented assemblies or missing sequence. Excessively high coverage may increase computational demands and can complicate assembly when reads contain errors, duplicated fragments, or multiple closely related genomic copies. Coverage requirements depend on genome size, sequencing technology, heterozygosity, repeat content, and the assembly method.
- Sequencing errors can introduce additional problems. Substitutions, insertions, and deletions may create false differences between reads, making it difficult to identify correct overlaps and genomic connections. High error rates can produce incorrect branches in assembly graphs, fragmented contigs, and inaccurate consensus sequences. Read correction and genome polishing can reduce these errors, but they cannot always recover information that was not captured reliably during sequencing. Polishing also requires care because genuine biological variation should not be mistakenly removed.
- GC-content bias can affect sequencing coverage and assembly representation. Genomic regions with very high or very low GC content may be underrepresented because of library preparation, amplification, sequencing chemistry, or other technical effects. These regions may receive insufficient coverage and become difficult to assemble. As a result, some genes and genomic regions may be missing even when the overall sequencing depth appears adequate. Coverage distribution and GC-content profiles should therefore be examined during assembly quality assessment.
- Chimeric reads and contamination can also affect genome assembly. A chimeric read contains sequence fragments originating from different genomic regions or molecules, which may lead to incorrect connections during assembly. Contamination introduces DNA from unintended organisms or sources, potentially creating additional contigs or misleading sequence joins. These problems can be investigated using read-quality assessment, taxonomic classification, coverage analysis, GC-content analysis, and alignment against suitable reference databases. Tools such as Kraken2, BlobToolKit, and CheckM2 can contribute to contamination screening in appropriate workflows.
- Assembly algorithms themselves impose limitations. Different assemblers use different approaches, including overlap-layout-consensus methods, de Bruijn graphs, string graphs, and other graph-based strategies. Each approach makes assumptions about sequencing errors, read coverage, genome complexity, and sequence relationships. An assembler may simplify an assembly graph by removing low-support paths or selecting a particular connection. Although this can improve continuity, it may also remove genuine biological variation or select an incorrect path in difficult regions. Assembly quality therefore depends not only on the sequencing data but also on algorithm design and parameter selection.
- Reference bias and assembly limitations can affect gene prediction and genome annotation. If a genomic region is missing, fragmented, or incorrectly reconstructed, gene prediction tools may fail to identify the complete gene structure. Exons may be missing, introns may be incorrectly represented, and duplicated genes may be collapsed or falsely expanded. Functional annotation can also be affected because missing or incorrect sequences may produce misleading similarity matches, protein predictions, or pathway assignments. Genome annotation should therefore be interpreted in relation to assembly quality.
- Population genomics and evolutionary studies are particularly vulnerable to reference bias. If a reference genome represents only one individual or population, genetic variants that are common in other populations may be underrepresented. This can affect estimates of genetic diversity, population structure, allele frequencies, and evolutionary relationships. In animal and plant breeding, a reference genome from one breed or variety may not adequately represent other genetic backgrounds. The use of multiple reference genomes, pangenome resources, and population-specific reference panels can help reduce these limitations.
- Pangenome analysis provides an important approach for addressing the limitations of a single linear reference genome. A pangenome represents the genetic diversity found across multiple individuals or strains of a species. It can include a core genome shared by most or all sampled individuals and an accessory genome containing sequences present in only some individuals or populations. Graph-based pangenome representations allow alternative sequences and structural differences to be represented within a shared framework. Resources and tools such as the Human Pangenome Reference Consortium, Minigraph-Cactus, and other graph-based approaches support the development and analysis of pangenome references. Pangenomes can improve the representation of genetic diversity, although their usefulness depends on the diversity and quality of the genomes included.
- Graph-based genome representations are also useful for reducing dependence on a single reference sequence. A traditional linear reference provides one primary path through the genome, whereas a genome graph can represent alternative paths corresponding to different alleles, haplotypes, or structural variants. Reads can be aligned to graph representations that contain more of the known sequence diversity, potentially reducing reference bias in variant discovery and genotyping. However, graph-based methods introduce additional computational and analytical complexity, and their performance depends on graph construction, representation, and the availability of suitable population data.
- Several strategies can help reduce reference bias and improve assembly reliability. One approach is to use a reference genome that is closely related to the target organism and has high-quality sequence and annotation. Another is to perform de novo assembly when reference divergence or structural variation is expected to be substantial. Combining short- and long-read sequencing data can improve both nucleotide accuracy and sequence continuity. Long-read sequencing is especially useful for resolving repetitive regions and large structural variants, while high-accuracy short reads or high-fidelity long reads can support consensus correction.
- Alternative reference genomes and pangenome resources can also improve representation of genetic diversity. In population studies, researchers may use multiple reference genomes, graph-based references, or population-specific reference panels to reduce the preference for one reference sequence. In transcriptomic and allele-specific expression studies, bias-aware mapping methods can be used to reduce unequal alignment of alternative alleles. One approach is to construct personalized or diploid references that represent both alleles, while other approaches use variant-aware or graph-based alignment.
- Variant-aware mapping is particularly useful when known genetic variants are available. Instead of aligning reads only to a single reference sequence, the analysis can account for alternative alleles at known variant positions. This may reduce the loss of reads carrying non-reference alleles. However, variant-aware methods cannot fully correct bias caused by unknown variants, missing structural sequences, or incomplete reference representation. Their effectiveness depends on the quality and completeness of the available variant information.
- Genome assembly limitations can be evaluated using complementary quality assessment tools. QUAST reports assembly statistics and supports reference-based comparisons that can reveal possible structural discrepancies. BUSCO evaluates the recovery of conserved single-copy orthologs and provides an estimate of gene-space completeness. Merqury uses k-mer comparisons to assess consensus accuracy and completeness-related properties. CheckM2 estimates completeness and contamination in microbial genomes, while BlobToolKit helps investigate taxonomic composition, GC content, and coverage patterns. These tools provide different types of evidence and should not be treated as interchangeable.
- The following command illustrates a reference-based assembly evaluation using QUAST:
quast.py assembly.fasta \
-r reference.fasta \
-o quast_results
- This analysis can identify differences between the assembly and the selected reference, including missing or rearranged regions. However, a discrepancy does not automatically represent an assembly error because the target genome may contain genuine biological differences from the reference.
- Minimap2 can be used to align long reads against an assembled genome to investigate whether the assembly is supported by the original sequencing data:
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 an appropriate alternative preset, such as map-hifi. The resulting alignments can be inspected using genome visualization software to identify regions with inconsistent read support, coverage changes, or potential structural problems.
- A BUSCO analysis can help evaluate whether expected conserved genes have been recovered:
busco \
-i assembly.fasta \
-m genome \
-l mammalia_odb \
-o busco_results \
-c 8
- The lineage dataset should be selected according to the organism being studied. A mammalian dataset is not appropriate for a plant, fungal, or microbial genome. BUSCO measures conserved gene recovery rather than complete nucleotide representation, so its results should be interpreted alongside other assembly metrics.
- Despite improvements in sequencing technologies and computational methods, complete genome reconstruction remains challenging for some organisms and genomic regions. Highly repetitive regions, centromeres, telomeres, segmental duplications, complex structural variants, and highly heterozygous sequences may remain unresolved. Some genomes also contain unusual chromosome structures, organellar sequences, or complex ploidy that require specialized assembly methods. Even chromosome-level assemblies may contain gaps, unresolved repeats, collapsed regions, or uncertain haplotype representation. Researchers should therefore avoid describing an assembly as complete solely because it has a high N50 or chromosome-scale scaffolds.
- For reliable and reproducible genome analysis, researchers should document the reference genome and version, sequencing platform, read type, sequencing depth, assembler and software version, and important computational parameters. They should also report whether the assembly was reference-guided or de novo, whether haplotypes were phased or collapsed, and which quality assessment methods were used. Any suspected contamination, reference discrepancies, or unresolved regions should be described transparently. Preserving original reads, assembly graphs, intermediate files, and analysis reports makes it possible to investigate errors and reproduce the assembly workflow.
- Reference bias and assembly limitations are fundamental considerations in modern bioinformatics. Reference bias can cause genuine genetic differences to be underrepresented when sequences are analysed against an unsuitable or incomplete reference, while assembly limitations can arise from repetitive DNA, heterozygosity, sequencing errors, uneven coverage, contamination, and algorithmic constraints. Strategies such as de novo assembly, long-read sequencing, hybrid approaches, pangenome references, graph-based alignment, and comprehensive quality assessment can reduce these problems. Nevertheless, no single approach eliminates every source of bias or uncertainty. Understanding these limitations is essential for interpreting genomic data accurately and producing reliable results in evolutionary biology, medicine, agriculture, population genetics, and comparative genomics.