- Variant representation across reference assemblies is an important concept in bioinformatics because genetic variants are described relative to a reference genome. A variant’s chromosome, genomic position, reference allele, and alternative allele depend on the particular reference assembly used to identify and report it. Consequently, the same biological variant may have different coordinates or allele representations in different genome assemblies. Understanding these differences is essential when comparing sequencing datasets, integrating genomic databases, performing variant annotation, interpreting clinical variants, and analysing genetic variation across populations.
- A reference genome assembly is a reconstructed representation of the DNA sequence of an organism. It provides a coordinate system against which sequencing reads are aligned and genetic variants are identified. Human reference genome assemblies include GRCh37, also known as hg19, and GRCh38, also known as hg38. Newer human reference resources, including the T2T-CHM13 assembly, provide additional sequence information and improved representation of genomic regions that were previously incomplete or difficult to assemble. Other organisms also have multiple reference assemblies, often representing different individuals, strains, breeds, or versions of a species’ genome.
- A variant is generally represented by its chromosome or contig, genomic position, reference allele, and alternative allele. In VCF format, these correspond to the CHROM, POS, REF, and ALT fields. For example, a single-nucleotide variant may be represented as chromosome 1, position 100000, reference allele A, and alternative allele G. This description is meaningful only when the reference assembly is known. If the same genomic region is represented differently in another assembly, the variant may have a different coordinate or surrounding sequence.
- Reference assemblies are not simply identical sequences with different labels. New assemblies may incorporate corrections to sequencing errors, improved contig ordering and orientation, additional genomic regions, updated gap representations, and better resolution of repetitive or structurally complex regions. They may also differ in chromosome sequences, alternate loci, unplaced contigs, and the representation of highly variable genomic regions. As a result, genomic coordinates cannot always be transferred directly from one assembly to another by assuming that the same position refers to the same biological location.
- For example, a variant reported at position 100000 in GRCh37 may correspond to a different position in GRCh38. The difference may be small in a region that has undergone a simple sequence correction, or more substantial where insertions, deletions, rearrangements, or changes in contig structure have occurred. Some regions may not have a clear corresponding location in the target assembly. Therefore, a variant should always be associated with its reference assembly and, where possible, the exact reference sequence and version used during analysis.
- The same biological variant can have different VCF representations across assemblies because both the genomic coordinate and the reference sequence may change. The REF allele in a VCF record must match the sequence at the specified position in the selected reference assembly. If the reference sequence differs between assemblies, the reference and alternative allele strings may need to be updated to describe the same biological change. Insertions and deletions may also require different coordinates and allele representations because the surrounding sequence context has changed.
- Variant representation across assemblies is therefore more complex than changing the chromosome position. A correct conversion must preserve the underlying biological sequence change while adapting its representation to the target reference genome. This requires identifying the corresponding genomic region, determining whether the variant can be mapped unambiguously, checking the target reference allele, and normalizing the resulting representation according to the target sequence.
- The process of converting genomic coordinates or variant locations from one reference assembly to another is commonly called liftover. Liftover tools use mappings between genomic regions in two assemblies to identify the corresponding target coordinates. These mappings are often based on pairwise sequence alignment or precomputed assembly-to-assembly alignment files. The conversion may be straightforward in well-conserved regions, but it can be difficult in repetitive sequences, duplicated regions, segmental duplications, rearranged regions, and sequences that are present in only one assembly.
- A common approach to human genome liftover uses chain files. A chain file describes aligned regions between two genome assemblies, including the positions and orientation of corresponding sequences. Tools use these mappings to convert coordinates from a source assembly to a target assembly. The UCSC Genome Browser provides the liftOver utility and chain files for supported assembly conversions. CrossMap is another widely used tool for converting genomic coordinates and related annotation files. Picard also provides a LiftoverVcf tool specifically designed to convert VCF records between reference assemblies using a chain file.
- The following example illustrates coordinate liftover using the UCSC command-line utility:
# Convert BED coordinates from GRCh37 to GRCh38
liftOver \
input.bed \
hg19ToHg38.over.chain.gz \
output.hg38.bed \
unmapped.bed
- This command uses a chain file to map genomic intervals from the source assembly to the target assembly. The output file contains successfully converted intervals, while the unmapped file records intervals that could not be converted. The exact chain file must correspond to the intended source and target assemblies. BED coordinates use a zero-based, half-open convention, which differs from the one-based position convention used by VCF. This distinction is important when converting between file formats and interpreting coordinates.
- For VCF files, Picard’s LiftoverVcf can be used to convert variant records between reference assemblies:
picard LiftoverVcf \
I=input.vcf.gz \
O=output.hg38.vcf.gz \
CHAIN=hg19ToHg38.over.chain.gz \
REJECT=unmapped_or_rejected.vcf.gz \
R=reference_GRCh38.fa
- The input VCF must be associated with the source assembly, while the R option specifies the target reference genome. The chain file describes the mapping between the assemblies. Successfully converted records are written to the output file, while records that cannot be converted or fail relevant checks are written to the reject file. Picard can also account for changes in strand orientation and reference allele compatibility according to its implementation. The output should still be validated and normalized against the target reference.
- CrossMap provides another option for converting VCF records:
# Example: convert VCF coordinates using CrossMap
CrossMap.py vcf \
hg19ToHg38.over.chain.gz \
input.vcf \
reference_GRCh38.fa \
output.hg38.vcf
- The exact command syntax and supported options depend on the installed CrossMap version. The reference FASTA must correspond to the target assembly. As with other liftover methods, converted records should be inspected for mapping failures, allele changes, and other representation issues.
- Liftover does not guarantee that every variant can be converted successfully. Some variants may fall within regions that are missing from the target assembly, have been extensively reorganized, or cannot be mapped uniquely. Other variants may map to multiple possible locations because of repetitive sequences or duplicated genomic regions. In such cases, a tool may reject the variant, mark it as ambiguous, or select a mapping according to its own rules. Researchers should retain information about rejected variants rather than silently discarding them.
- Strand orientation is another important consideration. A genomic region may be represented on the reverse strand in the target assembly relative to its orientation in the source assembly. When a variant is mapped across such a region, its alleles may need to be reverse-complemented. For example, the reverse complement of A is T, while the reverse complement of C is G. A source variant represented as A to G may therefore be represented as T to C after conversion to the opposite strand. This is a valid change in representation and does not necessarily indicate that the biological variant has changed.
- Reference allele validation is essential after liftover. The converted REF allele must match the target reference sequence at the new coordinate. If it does not, the record may have been mapped incorrectly, the reference assembly may be mismatched, the allele representation may need further processing, or the input data may contain an error. A reference mismatch should be investigated rather than automatically corrected without understanding its cause. The target reference FASTA should be indexed and used consistently throughout the conversion and validation workflow.
- After liftover, variants should generally be normalized against the target reference assembly. Normalization may include left alignment and trimming of insertions and deletions, as well as appropriate handling of multiallelic records. A variant that was left-aligned in the source assembly may not remain in the equivalent left-aligned representation in the target assembly because the surrounding reference sequence may differ. Therefore, normalization should be repeated after coordinate conversion rather than assuming that the source representation remains valid.
- BCFtools can be used to validate and normalize converted VCF records against the target reference:
# Normalize and validate variants against the target assembly
bcftools norm \
-f reference_GRCh38.fa \
-c w \
-Oz \
-o normalized.hg38.vcf.gz \
output.hg38.vcf.gz
# Index the normalized VCF
bcftools index normalized.hg38.vcf.gz
- The -f option supplies the target reference FASTA, and -c w reports reference allele mismatches as warnings. For strict workflows, mismatches should be investigated and handled according to documented quality-control rules. A stricter check can be performed using -c e, which causes the command to exit when a mismatch is encountered. The normalization process should be followed by inspection of the output and any warnings.
- Assembly differences can also affect variant annotation. Gene locations, transcript structures, exon boundaries, regulatory regions, and genomic features may change between reference assemblies. A variant that overlaps one gene or transcript in an older assembly may have a different annotation in a newer assembly. Annotation tools such as Ensembl Variant Effect Predictor (VEP), SnpEff, and ANNOVAR depend on the selected genome assembly and the corresponding annotation database. Therefore, annotation should be performed using resources that match the assembly to which the variant has been converted.
- Transcript identifiers and transcript versions are also important. A variant’s predicted consequence may depend on which transcript is selected, whether the transcript is protein-coding or noncoding, and which transcript annotation release is used. Even when the genomic variant is successfully converted between assemblies, the corresponding transcript and protein consequences may differ because gene models have been updated. Researchers should therefore document both the genome assembly and the transcript annotation release used in functional interpretation.
- Variant databases also use reference assemblies to describe genetic variation. Resources such as dbSNP, ClinVar, gnomAD, and the GWAS Catalog may provide coordinates or variant representations for particular genome builds. Some resources support multiple assemblies or provide tools for converting between them. However, database records may differ in their assembly coverage, allele representation, transcript annotation, and update history. A variant should not be matched to a database record using only its chromosome and position if the reference assembly and allele information are unknown.
- In clinical genomics, assembly consistency is particularly important because variant interpretation may influence diagnosis, treatment decisions, and genetic counselling. A variant reported using an older reference assembly may not match a database record that uses a newer assembly. Incorrect conversion can result in a wrong allele assignment, an incorrect gene annotation, or failure to identify relevant clinical evidence. Clinical workflows should preserve the original variant record, document the assembly conversion, validate the target reference allele, normalize the variant, and confirm its identity using appropriate genomic and transcript-level descriptions.
- The HGVS nomenclature system provides standardized descriptions of sequence variants at genomic, coding-DNA, and protein levels. HGVS descriptions are reference-sequence-dependent and should identify the sequence accession and version used. For example, a genomic variant description should be linked to a specific genomic reference sequence, while a coding-DNA description should identify the relevant transcript accession and version. This is important because a variant’s HGVS description may differ between reference assemblies or transcript versions even when it represents the same underlying biological change. Tools such as Mutalyzer and VariantValidator can help validate and convert variant descriptions, but their outputs must be interpreted in relation to the selected reference sequences and supported resources.
- A further challenge arises from alternate loci, haplotypes, and highly variable genomic regions. Some assemblies include alternate sequences representing alternative forms of particular genomic regions. These sequences can provide additional information about genetic diversity but complicate coordinate mapping and variant representation. The major histocompatibility complex (MHC), immunoglobulin regions, and other highly polymorphic or repetitive genomic regions are examples where a single linear reference may not represent all biological variation adequately. Variants in these regions may require specialized references, haplotype-aware approaches, or graph-based genome representations.
- The development of pangenome references provides an alternative to relying exclusively on a single linear reference sequence. A pangenome represents genomic diversity across multiple individuals or haplotypes and can reduce some forms of reference bias. Graph-based genome representations allow alternative sequences and paths to be represented within a shared structure. These approaches can improve the representation of population variation and complex genomic regions, although they introduce additional requirements for coordinate systems, variant representation, and interoperability with conventional VCF workflows. Traditional VCF coordinates may still be generated through projection onto a selected linear reference, but this projection can lose some information about the underlying graph path.
- Several quality-control steps should be performed when representing variants across reference assemblies. The source assembly and target assembly must be clearly identified, and the correct chain file and target reference sequence must be used. The number of successfully mapped and rejected variants should be recorded. Converted records should be checked for strand changes, reference allele mismatches, duplicate mappings, and unexpected coordinate shifts. Indels should be normalized against the target assembly, and multiallelic records should be handled consistently. The final file should be validated before annotation, merging, or downstream statistical analysis.
- For reproducible research, it is important to preserve the original variant file and the converted output. Researchers should record the assembly names and versions, chain file source, liftover software and version, command-line options, target reference FASTA, normalization procedure, and annotation database release. Checksums can be used to verify the integrity of reference files and analysis outputs. Rejected variants should be retained with their rejection reasons because they may contain biologically important information or reveal limitations in the assembly conversion process.
- Variant representation across reference assemblies is a fundamental issue in modern genomic analysis. A genetic variant is not fully described by its chromosome and position alone; its representation also depends on the reference sequence, allele orientation, normalization conventions, and annotation resources. Liftover tools such as UCSC liftOver, CrossMap, and Picard can convert variants between supported assemblies, but successful conversion requires careful validation. By combining accurate coordinate mapping, reference allele checking, strand correction, and normalization against the target assembly, researchers can improve the consistency and reliability of genomic data integration, variant annotation, clinical interpretation, and comparative analysis.