Different reference builds put the coordinates in different places

Your data gives one position for a variant; the source you looked it up in gives another. You run into this fairly early on.

The cause is usually that the reference genome builds do not match.

There are builds

The human reference sequence was not finished in one go. It has been revised. Two builds are in common use today.

Build	Also known as	Released
GRCh37	hg19	2009
GRCh38	hg38	2013

Revisions correct errors in the sequence and close gaps. As a result, the coordinates pointing at the same location shift. In some places by a few bases; in others considerably further.

The newer build is more accurate, but a great deal of material and literature built on the older one is still in circulation, and testing services differ in which they use. This is not a question of which is correct. It is a question of whether they match.

What to check first

A VCF file begins with a header. Reference information is often written there, so start by looking.

##fileformat=VCFv4.2
##reference=GRCh38
##contig=<ID=chr1,length=248956422>

Two places matter.

The ##reference line. If the build is written out, you are done quickly. Some files do not have it.

The length in ##contig. Chromosome lengths differ between builds, so you can work backwards from them. For chromosome 1, GRCh37 gives 249,250,621 and GRCh38 gives 248,956,422. Memorising these two numbers speeds up the check considerably.

With chr, or without

One more thing while you are here. Some files name chromosomes chr1; others just 1.

This is a separate matter from the build, but it is a common reason for a tool to quietly return “no match.” No error is raised — the result count simply comes back as zero, which makes it easy to miss. Check that the naming convention lines up, at the same time as the build.

Bringing the builds into line

When your data and the database you want to consult use different builds, there are tools to convert coordinates. Because the conversion is mechanical it invites complacency, and it deserves care.

Some coordinates cannot be converted. In regions whose structure changed in the newer build, there is no corresponding position to map to. The conversion tools drop those coordinates. Carry on without noticing and the variant is treated as “not found.”

Record the conversion. From which build to which, using which tool, at which version. Without that, you cannot reproduce the result when you come back to it later.

The practical conclusion

The most dependable approach is to anchor on the rsID.

Coordinates depend on the build; rsIDs do not. Rather than matching on coordinates, match on the rsID and then check the coordinate. Working in that order removes a large share of the losses caused by mismatched builds.

rsIDs have their own problems — they get merged, and they get withdrawn. That is a separate note.