Variant Coordinates and Allele Harmonization for Annotation

Why Coordinate Harmonization Comes First

A credible set identifies variants with posterior support for explaining a genetic association signal. To interpret those variants alongside functional tracks, each variant must be matched to the same genomic location and allele representation used by the annotation source. A failed match can mean the variant is absent from a track, but it can also mean that the two files describe the same biological site in incompatible ways.

The practical task is to determine what each file means by its chromosome, position, reference allele, alternate allele, strand, and variant representation. These fields are related, but none can safely be inferred from a variant label alone. In particular, a chromosome and position are meaningful only relative to a genome build and a coordinate convention.

Builds and Coordinate Conventions

A genome build assigns reference sequence and coordinates to chromosomes. The same biological region can have different coordinates in different builds because sequence content, gaps, and assembly decisions differ. Therefore, equal chromosome and position values in two files do not establish that the records refer to the same sequence. Conversely, a real variant can have different positions in two builds.

Coordinates also differ in their conventions. VCF records use 1-based positions: position 1 is the first base of a chromosome. BED intervals use 0-based, half-open coordinates: the start is included and the end is excluded. An interval with BED start \(s\) and end \(e\) covers the 1-based bases \(s+1\) through \(e\), inclusive. Thus a one-base BED interval \([s,s+1)\) refers to 1-based position \(s+1\), not \(s\).

This relationship follows directly from the conventions: BED counts the number of bases before its start coordinate, while a 1-based coordinate numbers the first base as 1. A BED interval from 0 to 1 contains the first base, whose 1-based position is 1; shifting both BED boundaries up by 1 gives the corresponding inclusive 1-based endpoints. For a one-base variant, checking this conversion explicitly prevents a frequent one-base error.

When a variant file and annotation file use different builds, conversion between builds is usually attempted with a mapping procedure such as liftOver. This is not a universal arithmetic offset. A source interval may map to a different position, map to more than one target interval, or fail to map. Even a successful coordinate conversion should be checked against the target reference sequence and allele representation. A build label should be recorded for each input, rather than assumed from a file's appearance or filename.

Reference and Alternate Alleles

For a small variant represented in VCF, REF is the reference sequence at the reported site and ALT is the alternate sequence. The reference allele is defined against a particular genome build and sequence orientation. It is not simply the allele that is common in a population, nor does REF imply the ancestral allele. A record whose REF does not match the specified reference sequence is a warning of a possible build, strand, position, or file-format problem.

For an allele-specific match, a useful key includes at least build, chromosome, normalized position, REF, and ALT. A locus-level track may report only an interval and thus support a positional overlap without distinguishing alleles. That kind of overlap is not evidence that the track refers specifically to the alternate allele. Keep the distinction clear: a position match can be adequate for a non-allele-specific interval query, but it is not an allele-resolved match.

The strand convention matters when alleles are reported relative to opposite DNA strands. To express an allele on the opposite strand, take its reverse complement: complement each base and reverse the sequence order. For a single-base substitution, A pairs with T and C pairs with G. Thus a C>T change on one strand corresponds to G>A when written on the opposite strand. The genomic position is unchanged by this orientation switch, provided the coordinate system itself has not changed.

Variant Representation and Normalization

A variant can have more than one textual representation, particularly an insertion or deletion in a repeated sequence. VCF uses an anchor base so that REF and ALT are nonempty. For example, an insertion may be encoded by including the base immediately before the inserted sequence. In a repeat, the same biological insertion can sometimes be placed at different nearby positions while preserving the same resulting sequence. A literal comparison of position and allele strings can therefore fail even when two records describe the same event.

Normalization applies consistent representation rules, commonly trimming redundant shared sequence from alleles and left-aligning indels as far as possible against the reference sequence. These operations require the correct reference build. Normalization does not make two different biological events equivalent; it gives equivalent representations a consistent form when the reference and rules agree. Multiallelic records also need care: splitting a record with multiple ALT alleles into separate variant records must preserve the correct REF, ALT, and any associated annotations.

For a VCF record with position \(p\) and REF length \(r\), the reference span occupies 1-based positions \(p\) through \(p+r-1\). For a simple substitution, this is one base. For an indel, the represented REF span and the biological insertion or deletion event are not interchangeable concepts. An annotation pipeline should specify whether it tests the VCF anchor, the REF span, or a normalized event position when querying intervals. In particular, an insertion occurs between reference bases; a one-base anchor overlap is a convention for matching, not proof that the inserted bases themselves occupy that reference position.

A Practical Harmonization Workflow

For credible-set annotation, harmonize records before interpreting whether a variant overlaps a functional track. The goal is not to manufacture a match. It is to determine whether the inputs make a match identifiable, and to retain an explicit unresolved status when they do not.

Diagnosing Failed and Misleading Overlaps

When a credible-set variant appears not to overlap a track, inspect the matching inputs in a fixed order. First compare build labels, then coordinate conventions, then chromosome and position, and finally alleles, strand, and indel representation. A useful record of the result includes the original values, harmonized values, conversion or normalization applied, and a status such as matched, not matched, or unresolved. This makes it possible to distinguish a genuine absence from a technical mismatch.

Harmonization also prevents false confidence. A position-only interval overlap does not establish that a track distinguishes the credible-set ALT allele from REF. Nor does successful coordinate conversion establish functional relevance. These checks answer whether the records can be compared at the intended level; they do not establish a molecular mechanism or show that a variant causes a trait.

• Wrong reasoning: “The chromosome and number look close, so the records must refer to the same variant.” Correction: Require compatible builds and coordinate conventions before comparing positions.
• Wrong reasoning: “The alleles differ, so the records cannot match.” Correction: Check whether one source reports the opposite strand, and reverse-complement only when supported by metadata.
• Wrong reasoning: “Different indel strings prove different events.” Correction: Normalize against the same reference before deciding whether representations are equivalent.
• Wrong reasoning: “The variant overlaps the track, so the track supports the ALT allele.” Correction: State whether the track is allele-specific; a position-only overlap usually does not resolve REF versus ALT.