How Do You Know When You've Looked Everywhere?

The statement that sounds like an answer but isn’t

“The genetic test came back.”

It is tempting to read that as a door closing — either it found something or it didn’t, and either way the matter is now settled. But once the architecture of these assays is understood, the statement turns out to mean something closer to “one drawer was checked.” There are many drawers. Each test opens exactly one of them and is largely blind to the rest. This is rarely spelled out, because it is inconvenient and a little unsettling, so it is worth setting down plainly — for anyone handed a result who is unsure whether it actually resolves anything.

When a structural change appears — say, an extra or missing copy of a stretch of DNA — the natural next question is whether the finding is the whole story, part of the story, or an incidental bystander caught on camera. Answering that does not mean running “a genetic test.” It means running a series of mechanistically distinct assays, each interrogating a different class of variation. The discipline lies in knowing which classes have been excluded and which have not yet been glanced at.

Different assays detect different classes of variation

There is no single “is the DNA intact” test. There are several, and their detection windows overlap far less than one might hope. It helps to think of DNA changes as different categories of edit in a very long document:

• A whole paragraph was duplicated or deleted. Large, structural, detectable if one is measuring dosage — copy number. The instrument is the chromosomal microarray (also reported as array-CGH or SNP array). It is typically run first, and it resolves gains and losses down to roughly tens of kilobases in targeted regions. But it measures copy number only. It cannot read sequence, it cannot detect balanced rearrangements (a translocation or inversion that moves material without changing the total dose), and it is blind to everything below.
• A single letter differs — a point change that alters one codon and flips the meaning. Dosage measurement cannot see this; the bases must actually be read. That is whole-exome sequencing (WES), which sequences the ~1–2% of the genome that codes for protein. Run as a trio — proband plus both parents — it gains substantial power, because it distinguishes inherited variants from those arising de novo, and de novo status in a constrained gene is a strong pathogenicity signal.
• A short motif was over-repeated — the same few bases tandemly expanded far beyond the normal range, like a key held down too long. This requires a dedicated repeat-expansion assay (the canonical example is Fragile X / FMR1 CGG-repeat testing, historically by repeat-primed PCR and Southern blot). Short-read sequencing routinely misses large expansions, because the reads are shorter than the repeat tract and cannot be uniquely mapped across it, unless a specialized expansion caller is explicitly applied.
• The bases are correct, but the wrong genes are switched on or off. The text is intact; the annotation is wrong. This is the regulatory chemistry layered on top of the sequence — DNA methylation and the associated chromatin marks that set each gene’s expression level — and it is read with methylation testing, commonly MS-MLPA. The same assay also flags uniparental disomy, in which both homologs of a chromosome derive from a single parent, disrupting imprinted loci. A pure sequence read sees nothing wrong here, because nothing is wrong with the sequence itself.
• The change is mosaic — present in some cells, absent in others. If a variant arose after the first cell division, a blood sample can be entirely clean while the variant resides in an unsampled tissue. Detecting this means sampling more than one tissue (a buccal swab, or banked fibroblasts cultured from a small skin biopsy) and, increasingly, sequencing to high depth so low-frequency alleles are not lost in the noise.

Five classes of variation. Five different instruments. And the operative fact is that each instrument is dark to the others. The dosage assay cannot resolve a point mutation. The sequence read cannot resolve an epigenetic misconfiguration. Running one, obtaining a clean result, and concluding “clean means all clear” is a category error. It means that one drawer was empty.

The ladder

The useful reframe is to stop treating this as a single test and treat it as a ladder. Each rung is a different class of variation, and the only honest question is: how many rungs have actually been stood on?

The bottom rung — large structural change, copy number — is the chromosomal microarray, which is usually where a copy-number finding surfaces in the first place. (A conventional karyotype sits alongside it for visualizing whole-chromosome aneuploidy and balanced rearrangements that an array, measuring only dose, will miss.)

The next rung is the one most often omitted: the regulatory layer — the switched-on-or-off chemistry — read with methylation testing (MS-MLPA). A microarray physically cannot resolve it. Not “probably didn’t” — cannot; it is the wrong measurement. It is comparatively inexpensive and specific, and it is the rung most frequently skipped. Above it sits the repeat-expansion / Fragile X (FMR1) assay, then the slow, comprehensive read-every-coding-letter test — whole-exome sequencing (WES), then whole-genome sequencing (WGS) to reach the noncoding regulatory regions, deep-intronic splice variants, and structural breakpoints the exome skips. Reserved for the residual cases are the assays for rare structural mechanisms: long-read sequencing (which spans repeats and resolves complex rearrangements short reads cannot phase), optical genome mapping (which images very large structural and balanced events at megabase scale), and mitochondrial DNA sequencing (a separate genome, with heteroplasmy that nuclear assays do not assess).

The point worth holding onto: a workup can be much further along than “we did a genetic test” implies, and much less finished than “the test was negative” implies. Both statements are traps. The reality is a checklist — and a checklist is something that can be worked through one rung at a time.

Three reads of the same genome: exome, whole-genome, long-read

The three sequencing assays on that ladder are easy to conflate, because all three “read the DNA.” They differ in how much they read and how they read it, and those two differences decide what each one can and cannot find.

Whole-exome sequencing (WES) reads only the exome — the ~1–2% of the genome that codes for protein. The reasoning is economic and statistical: most variants with a known, large effect on disease fall in coding regions, so sequencing only those regions concentrates effort (and read depth) where the diagnostic yield per base is highest. The cost is everything it skips: the other ~98% of the genome — promoters, enhancers, deep-intronic splice sites, and structural breakpoints that land between genes — is simply not in the data. A causal variant sitting in a regulatory region is not “missed” by WES so much as never sampled.

Whole-genome sequencing (WGS) reads the entire genome, coding and noncoding alike, at a more uniform depth. The practical advantages over WES are threefold. First, it sees the noncoding ~98% — regulatory and deep-intronic variation an exome cannot reach. Second, it covers the coding regions more evenly: WES relies on a capture step that enriches for exons, and that step leaves coverage gaps and GC-biased dropouts, so WGS often resolves exonic variants that an exome read poorly despite “covering” them on paper. Third, its even, genome-wide depth makes it markedly better at calling copy-number and structural variants from sequence alone — it can both confirm a microarray finding and define its breakpoints to the base. The trade-offs are cost, a larger data and storage footprint, and a heavier interpretive burden: reading the whole genome means confronting far more variants of uncertain significance, including in noncoding regions whose grammar is still poorly understood.

Both WES and WGS, in their standard form, are short-read technologies: the DNA is fragmented into pieces of a few hundred bases, each read separately, then the fragments are computationally mapped back to a reference. That fragmentation is the hidden limitation. Where the genome is repetitive — long tandem repeats, segmental duplications, the recurrent breakpoint regions that mediate copy-number variants — a short read cannot be placed uniquely, because the same short sequence appears in many places. Large repeat expansions are longer than the read itself and collapse into ambiguity. And because each fragment is read in isolation, short reads struggle to phase variants — to determine which ones sit together on the same physical copy of a chromosome.

Long-read sequencing (the third assay) reads single molecules tens of thousands of bases long, end to end, with no fragmentation. That single change resolves exactly the cases short reads cannot: it spans repeat expansions and reads straight through them, it maps cleanly across segmental duplications and the repeat-flanked regions where copy-number variants recur, it resolves balanced rearrangements (inversions and translocations that move sequence without changing dose, invisible to a dosage assay and easy for short reads to miss at the junction), and it phases long stretches natively — assigning variants to the maternal or paternal copy, and detecting methylation directly off the same molecule on some platforms. Its historical costs were a higher per-base error rate and price, both of which have fallen substantially. In a completeness framework, long-read sequencing is the rung reserved for when a structural or repeat mechanism is suspected but short-read WGS has come back unrevealing — it reaches the parts of the genome the short-read assays are constitutively blind to.

The compact hierarchy: WES reads the coding 1–2%. WGS reads all of it, more evenly, and calls structure better. Long-read reads it in long, phaseable, repeat-spanning stretches. Each step widens what is visible, and each adds cost and interpretive load. None of them, individually, reads the methylation layer or guarantees detection of low-level mosaicism — which is why sequencing depth is necessary but never sufficient on its own.

	Exome (WES)	Whole-genome (WGS)	Long-read
Reads	Coding ~1–2%	Entire genome (~100%)	Entire genome, in long molecules
Read length	Short (~150 bp)	Short (~150 bp)	Long (10,000s of bp)
Point mutations	Yes, in coding regions	Yes, genome-wide	Yes
Noncoding / regulatory	No	Yes	Yes
Copy-number / structural	Weak	Good, breakpoints to the base	Best
Repeat expansions	No	Limited	Yes — reads through them
Balanced rearrangements	No	Partial	Yes
Phasing (which parent)	No	Limited	Yes, natively
Methylation	No	No	Yes, on some platforms
Cost / interpretive load	Lowest	Higher	Highest

The one-line rule of thumb: if the exome is the right tool when you suspect a typo in a known gene, WGS is what you reach for when the answer might be hiding in the noncoding 98% or in a structural change — and long-read is the last resort for repeats and rearrangements the short-read assays are physically blind to.

Why a negative result is not the closure it appears to be

Suppose the comprehensive sequence read comes back clean — trio whole-exome sequencing (WES) on the proband and both parents. The instinct is relief, followed almost immediately by a sharper unease: if it is clean, what is the explanation?

This is the intuition worth retraining. A negative result on one rung does not lower the prior probability that something is there. It redistributes the remaining probability onto the rungs that have not been climbed. A clean exome does not assert “there is nothing.” It asserts “it is not a coding point variant in a known gene — so the remaining probability now sits on methylation, repeat expansions, noncoding and structural variation that a whole-genome read would reach, and mosaicism a single tissue would miss.” A negative does not close the search. It aims it.

There is also a humbler possibility worth room for. Many structural findings are incompletely penetrant and variably expressive — in most individuals who carry them, they produce little or no phenotype. The same copy-number variant can be pathogenic in one carrier and silent in another, depending on the variant on the second allele, modifier loci, and stochastic developmental factors. A finding can therefore be one contributor among many small ones rather than a sole cause, with no single villain to indict. That is not a satisfying resolution. It can still be the correct one. Doing this rigorously means being willing to hold an explanation that does not collapse to a single clean cause.

The ceiling is below one hundred percent

It is worth being explicit about the ceiling, because the pull toward certainty is strong and the data do not support it.

One can climb a long way — a handful of assays, two of them inexpensive, covers most of the ladder. But the final stretch never fully closes, for three honest reasons:

Not every disease gene is known yet · the hardest regions of the genome still cannot be read cleanly · some phenotypes are distributed across hundreds of small contributions, not one

The first reason keeps a peculiar door open. A variant of uncertain significance today — a result no one can yet interpret — can become a diagnosis in a few years as gene–disease relationships are established. Stored raw data are re-interpretable later. So one of the highest-yield actions is not a new assay at all: it is retaining the raw sequence files and requesting periodic reanalysis — typically every one to two years — against updated databases as the field accumulates knowledge.

There is something genuinely useful in that. Not “everything will be known,” but “today’s data can keep working years from now without being regenerated.”

What to do with all of this

The shift that helps is to stop asking “is the test back yet” as though one envelope ends the story, and instead maintain a single page — a ladder with checkboxes. Which rung is done, which is next, where it is performed, and precisely what it can and cannot resolve. When a result lands, the question is not “good or bad.” It is “which rung did this clear, and which rung is now the most important thing no one has examined.”

That substitution converts an open-ended unknown into a finite task list. The unknown has no edges. A task list does.

One may never learn which rung holds the answer, or whether there is a single answer to hold. But one can know exactly which drawers have been opened and which have not — and that no one gets to quietly declare the matter finished while drawers remain shut. Some days that is the most the science can offer. Most days, it is enough.