Sorting "nothing came up" into three

The rsID you were looking for is absent from your results. Was it the same as the reference, was the position never read, or is it a kind this method cannot read at all? All three look like "no match" and none of them mean the same thing.

You search your results for the rsID you are after and nothing comes back. This happens constantly once you are working through your own data.

There are two ways to read it, and which way you go inverts the conclusion.

The file lists only the places that differed

A VCF is a record of differences from the reference sequence. Where your sequence matched, as a rule no line is written.

So “not in the file” can mean either of these:

The position matched the reference, so no line exists
The position was never read well enough for anything to be decided

The file format renders both as the same absence. That is where this starts.

There is also a format that writes out everything that was read, covered positions included — a gVCF. Check which one you are holding before anything else. If it is the differences-only kind, the rest of this applies.

Avoid both short cuts, in both directions

Two mistakes are common here, and they point opposite ways.

✗ Absent means no change
✗ Nothing nearby means the position was not read

The first hardly needs arguing: it writes down “normal” for a place that was never read.

The second is the one that slips past. Variants are sparse to begin with. A single genome differs from the reference at a few million positions, which, spread across the length of the genome, works out to something like one in several hundred to a thousand bases. Finding nothing immediately beside a given position is entirely unremarkable.

Read that as “not covered” and you accumulate a pile of unresolved entries. A record that is mostly unresolved says very little more than a blank page.

Fix the position first, then look at it

Work position by position, not gene by gene.

Establish the GRCh38 coordinate for the variant (look it up from the rsID)
Check whether anything was called in the immediate vicinity of that coordinate
If so, the region was read
If not, widen the window and look for evidence that the region was read at all
If widening still turns up nothing, put the position on the unread side

How far to widen has no objective default. Which is exactly why you must write down the width you chose and hold to it. Widen or narrow it position by position and your criterion starts moving to suit what you were hoping to find.

When something was called nearby, record that too. Whether you can write “there are calls near this position, so the region was read” is what decides how much the record is worth later on.

The third box — what this method cannot read

There is a category that is not quite “unread” either.

Differences in copy number, including whether a gene is present at all
Places where the number of repeats is itself the polymorphism
Regions with close copies elsewhere in the genome, where a fragment’s origin cannot be pinned down
Chromosome ends, and regions the sex chromosomes hold in common

None of these yield to reading harder. They sit outside what a method built on lining up short fragments can do. Settling them takes a different approach.

Write the three differently

Sorted out, the three become three sentences.

“Matched the reference” — with what established it
“Not read” — with which window came back empty
“Cannot be read this way” — with what it would take to settle

Do not collapse the three into one “no match.” The moment you do, whoever reads the table next cannot tell them apart — and whoever reads it next is usually you.

Unresolved is not a safe default

One caution pointing the other way, to close.

Mark everything unresolved out of caution and the record becomes unusable in a different manner. Where something was read, say it was read. If working through the procedure lets you state that a position matched the reference, you are entitled to state it.