The single most important habit in genomic medicine is to treat a published yield figure as a property of a cohort, not of a test. The same platform — say, trio WES — is reported at 10% in one paper and 72% in another. Neither number is wrong; they describe different populations. A yield is a fraction, and almost all of the variation lives in the denominator: who was selected, what was already ruled out, and how the genome was interrogated. Six variables account for nearly all of the spread.

Cohort selection — the dominant force. Diagnostic yield rises with the prior probability of a monogenic cause. A specialty leukodystrophy clinic enrolls children with an MRI pattern that already implies a Mendelian white-matter disease, so almost everyone in the denominator is genetically diagnosable — yields of 40–72% follow naturally. A population-level developmental-delay cohort sweeps in mild, multifactorial, and environmentally influenced phenotypes, diluting the Mendelian fraction to 15–25%. The test never changed; the pre-test probability did. This is why the first question about any yield figure is never "which platform?" but "who was in the denominator?"

Singleton vs. trio. Trio sequencing (proband + both parents) approximately doubles yield for de novo-enriched phenotypes (OR ~2.04, Clark et al. 2018). The mechanism is not deeper sequencing — it is interpretive power. Parental data lets the lab instantly flag a variant as de novo (absent in both parents), the genetic signature of most severe early-onset disease, and phase compound heterozygous variants to confirm they sit on opposite alleles. A singleton finds the same variants but cannot resolve their significance, so many land in the VUS pile. For severe DEE, GDD, or MCA, trio is essential.

Prior testing erodes the denominator. A post-CMA-negative WES cohort yields ~25–35%, while first-tier WES in the same phenotype yields ~30–45%. The CMA already removed the CNV-driven diagnoses, so WES is now working a harder, pre-filtered population. Sequential testing strategies must be read in this light: each negative test enriches the remainder for whatever the next test detects.

Database maturation makes yield dynamic. Reanalysis of older WES data yields new diagnoses in ~10–25% of unsolved cases — not because the sequence changed, but because the gene-disease knowledge base did. New gene-disease associations are validated continuously, so a variant that was uninterpretable in 2021 may be diagnostic in 2024 from the identical FASTQ file. A negative WES is therefore time-stamped, not permanent.

CMA platform. SNP arrays detect UPD and long stretches of autozygosity (AOH); oligo-only arrays miss both. This is not a minor footnote — UPD is the mechanism behind a subset of imprinting disorders (Prader-Willi, Angelman), and AOH flags consanguinity that raises the prior for recessive disease and points the analysis toward homozygous regions.

Severity and specificity. Younger onset, multi-system involvement, comorbid epilepsy, and distinctive examination or imaging features each independently raise yield, because each shifts the phenotype away from the common, multifactorial middle and toward the rare, single-gene tail. Mild isolated phenotypes (isolated ASD without ID) sit squarely in that multifactorial middle and consistently show the lowest yields.

The three first-line genomic tests are not a quality ladder where more sequence is simply better — they are different windows onto the genome, each blind to what the others see best. Choosing well means matching the window to the kind of variant the phenotype predicts.

Chromosomal Microarray (CMA) — Pooled NDD yield: ~10%

CMA measures dosage: it counts copies of DNA across the genome and flags deletions and duplications down to ~50–200 kb, plus whole-chromosome aneuploidy. SNP arrays add a second dimension — genotype — letting them detect UPD and autozygosity that copy-number-only arrays cannot. What CMA fundamentally cannot do is read sequence: a single-base change that destroys a gene is invisible because the amount of DNA is unchanged. So CMA misses SNVs, small indels, balanced rearrangements (no net dosage change), and repeat expansions. Its enduring value is being cheap, fast, decades-validated, and the most sensitive tool for the CNV space — which is why it stays first-tier for MCA, IESS, and ID, where copy-number disease is common.

Whole Exome Sequencing (WES) — Pooled NDD yield: ~36%

WES reads the ~2% of the genome that codes for protein, where the large majority of known Mendelian disease lives. It excels at the variant class CMA is blind to — SNVs and small indels — which inverts CMA's blind spot and explains why the two are complementary rather than redundant. Its own blind spots follow directly from what it captures: it ignores deep intronic and regulatory variants outside the captured exons, calls CNVs less reliably than CMA (uneven capture depth degrades dosage inference), resolves balanced SVs poorly, and does not read repeat expansions. It is the standard first-tier test for undiagnosed NDD/ID at most academic centres.

Whole Genome Sequencing (WGS) — Pooled NDD yield: ~41%

WGS removes the capture step and sequences the whole genome at uniform depth. In principle it sees everything WES sees plus deep intronic/regulatory variants, balanced SVs, better CNV breakpoints, and improved mitochondrial coverage; modern pipelines (e.g., ExpansionHunter) can even screen some STR loci, though sensitivity is variable. In practice, the headline result is sobering: WGS yield is NOT significantly higher than WES overall (OR 1.13, p=0.50). The reason is that most currently interpretable disease variants are coding, so the extra non-coding territory WGS reads is largely uninterpretable today. The incremental gain is therefore concentrated where non-coding or structural pathology is enriched: post-WES-negative patients, leukodystrophies, and atypical CP.

The repeat-expansion blind spot. Short-read sequencing assembles the genome from ~150-bp fragments, which cannot span a long, monotonous tandem repeat — the reads simply pile up ambiguously, so the expansion is unmeasurable. This is a structural limitation of the chemistry, not a coverage problem, and it means standard WES (and largely short-read WGS) miss the most common repeat-expansion disorders: Friedreich ataxia, the SCAs, CANVAS, Fragile X/FXTAS, DM1/DM2, Huntington, and C9orf72. These require dedicated repeat-primed PCR, Southern blot, or long-read sequencing. Before declaring a genomic workup complete, always ask whether the phenotype points to a repeat disorder that the sequencing could never have seen.

Within epilepsy, yield tracks a single organising principle: the more the seizures index brain that was built wrong rather than brain that fires wrong, the higher the monogenic fraction. Encephalopathic, early-onset, developmentally regressive epilepsies sit at the high-yield end; isolated seizures in a normally developing child sit at the low-yield end. The clinical sub-phenotype, more than the test, predicts the number.

NICU / Neonatal Encephalopathy

• rWES 20–35%; rWGS 35–50% (in the GEMINI randomized trial, rWGS yielded 49% vs. 27% for a targeted neonatal panel — Maron et al. 2023)
• ~18% of NICU admissions carry a Mendelian disease
• Diagnosis changed management in 38–50% of cases — the highest clinical utility in paediatric genetics

The neonatal brain has limited ways to express distress, so a wide range of monogenic disorders converge on the same encephalopathic picture — making the phenotype non-specific but the underlying genetic causes abundant. That combination, plus a narrow therapeutic window where a diagnosis can redirect care within days, is why speed (rWGS/rWES returning in days, not weeks) is itself part of the clinical utility here.

Developmental & Epileptic Encephalopathy (DEE)

• WES 24–40%, WGS 35–50%
• Highest yields in specific syndromes: EIMFS/Dravet ~78%, early infantile DEE ~43%

The syndrome-specific spread is instructive: a sharply defined electroclinical syndrome (EIMFS, Dravet) behaves like a specialty cohort in miniature — the phenotype is so constrained that nearly everyone has a detectable variant, hence ~78%. Looser "DEE" labels pull in more heterogeneity and lower the number.

Infantile Epileptic Spasms (IESS)

• Meta-analysis: CMA 14%, WES 26% (CI 21–31%)
• Genetic diagnosis enables precision therapy in 61.6% of genetically explained cases

IESS is the phenotype where CMA earns its first-tier place even in the sequencing era — a 14% CNV yield is high enough that ordering CMA alongside WES, not after it, is justified.

Non-DEE Epilepsy (focal or generalized, normal development)

• WES 10–18%; gene panel is a reasonable first step
• CMA rarely diagnostic for point-mutation epilepsies (SCN1A, GABRA1, GABRG2)

Here normal development signals brain that fires wrong, not brain built wrong: the monogenic fraction is small and skewed toward ion-channel point mutations, so a focused panel is an efficient first pass and CMA contributes little.

Drug-Resistant Epilepsy

• Drug resistance independently predicts higher yield — it is a clinical marker that the epilepsy is unlikely to be benign or self-limiting, enriching for monogenic causes
• All-epilepsy meta-analysis (Sheidley et al. 2022, n=39,094): CMA 9%, WES 24%, WGS 48%

Across neurodevelopmental phenotypes, two levers move yield: how severe and syndromic the presentation is, and how much family structure the lab is given to interpret the data. Severity raises the monogenic prior; trio structure converts found variants into diagnoses. The phenotypes below are arranged roughly from high to low monogenic fraction, and the reasoning matters more than any single number.

GDD (infant/toddler, unexplained)

• WES first-tier 30–40%; trio + CNV-seq up to 61% (Zhang 2024, n=434)
• Trio ~doubles yield vs. singleton (OR ~2.04)
• Strongest yield predictors: moderate-to-severe impairment, age 12–24 mo, craniofacial features

The jump from ~35% singleton WES to 61% with trio + CNV-seq is the whole argument for first-tier trio in one cohort: severe early GDD is enriched for de novo SNVs and de novo CNVs, so pairing trio interpretation with parallel copy-number detection captures both de novo mechanisms at once rather than missing one.

Intellectual Disability (child/adolescent)

• WES 30–45%; WGS 35–50%
• ACMG 2021 supports WES/WGS as first-/second-tier (Manickam et al. 2021)
• DDD study (Wright et al. 2023): 41% yield (exome + microarray)

The ACMG shift to WES/WGS as a first-tier test reflects accumulated evidence that exome/genome sequencing out-yields the older stepwise (karyotype → CMA → single-gene) pathway while reaching diagnosis faster — sequential testing mostly delays the answer.

Isolated ASD (without ID, epilepsy, or distinctive features)

• Lowest NDD yield: WES 10–15%
• Comorbid ID raises yield to ~25–30%
• CMA retains value for recurrence counseling even without diagnosis

Isolated ASD is the canonical low-yield, highly polygenic phenotype: the more it is just ASD, the more its architecture is common-variant and environmental rather than monogenic. Each added feature — ID, epilepsy, distinctive features — pulls it back toward the Mendelian tail and roughly doubles the yield.

Multiple Congenital Anomalies (MCA)

• CMA is productive first step at 15–25%
• WES/WGS 35–55%; MCA+ID up to 62%
• Consanguinity substantially raises recessive diagnosis yield

Multiple malformations imply a disturbance early in development affecting several organ fields — frequently a contiguous-gene deletion or aneuploidy, which is exactly the CNV space CMA reads best, justifying CMA first. Consanguinity then shifts the prior toward homozygous recessive disease, and SNP-array AOH data directs the sequencing analysis to those regions.

Cerebral Palsy (unexplained, no clear perinatal cause)

• WES overall 31% (CI 24–39%), pediatric-specific 35%, strict exclusion 42% (Gonzalez-Mantilla 2023)
• CP+ID yields 38% vs. 18% without ID

CP is the phenotype whose yield rises as you subtract patients: the stricter the exclusion of perinatal and structural causes, the more the remaining "unexplained" denominator is enriched for genetic disease — 31% becomes 42%. This is the denominator principle running in reverse, and it reframes unexplained CP from a static-injury label into an actionable genetic diagnosis.

Macrocephaly

• WES 20–40%; key genes: PTEN, PIK3CA, NF1, RAS-MAPK pathway
• Somatic mosaic overgrowth (PIK3CA, AKT3) may need deep sequencing or tissue biopsy

Overgrowth phenotypes carry a mechanistic trap: when the driver is a post-zygotic somatic variant, it may be absent or vanishingly rare in blood and present only in affected tissue. Standard blood WES can return falsely negative, so deep sequencing or a tissue biopsy of the overgrown region may be required to find a variant that is real but not in the sampled cells.

Movement and white-matter disorders make the central lesson of this module concrete: the molecular architecture of a phenotype decides which test can ever find the answer. Ataxia is split down the middle — one half is point mutations that sequencing reads cleanly, the other half is repeat expansions that sequencing structurally cannot see. Leukodystrophy, by contrast, is where front-loaded phenotyping (the MRI) pushes yield to the top of all neurogenetics.

Episodic Ataxia

• WES 20–35%; almost exclusively ion-channel point mutations (KCNA1, CACNA1A, CACNB4, SLC1A3)
• Gene panel competitive with WES for well-defined phenotype

Because episodic ataxia is mechanistically a channelopathy — paroxysmal dysfunction of otherwise normal cerebellum — its variants are SNVs that both panels and WES detect well, so a tightly defined phenotype gains little from the broader exome.

Progressive (Hereditary) Ataxia

• WES ceiling ~50% in specialty cohorts (Fogel et al. 2020); WGS 40–55%
• Most common hereditary ataxias are NOT detected by WES/WGS — they require dedicated repeat testing:
• Friedreich ataxia (FXN GAA repeat) → repeat-primed PCR
• SCA types (CAG repeats) → repeat analysis
• CANVAS (RFC1 AAGGG repeat) → Southern blot or long-read sequencing
• FXTAS (FMR1 CGG premutation) → FMR1 PCR
• Always ask after a negative WES in ataxia: has dedicated repeat expansion testing been sent?

The ~50% WES "ceiling" is not a technical failure to be solved by sequencing harder — it is the boundary of the variant class WES can read. The unsolved half is disproportionately repeat-expansion disease, so a negative WES in progressive ataxia is a positive prompt to send dedicated repeat testing, not evidence against a genetic cause. CANVAS (ataxia + sensory neuropathy + cough) is the trap case: a textbook-fit phenotype that short-read sequencing will report as negative every time.

Leukodystrophy (MRI-selected, genetic suspected)

• WES 50–72% — among the highest yields in all of neurogenetics
• WGS 72–90%+ in dedicated programmes (Zerem et al. 2023: 89.6% with all modalities)
• GWMD cohort: 72% overall, 77% for onset <3 yr, 85% for hypomyelination
• MRI pattern recognition is the essential pre-test step

Leukodystrophy proves that yield is engineered before the sample is drawn. The MRI is a free, high-resolution phenotyping step: categorising the white-matter signature (hypomyelination vs. demyelination vs. cystic vs. vacuolating) collapses a vast differential into a handful of candidate disorders, so the sequencing arrives with a near-monogenic prior already established — exactly the specialty-cohort effect from section one, generated by imaging rather than referral. The same logic explains the internal gradient: hypomyelination (85%) is a more constraining MRI pattern than "leukodystrophy" in general (72%), so it carries the higher yield. See the Hereditary Ataxias module for detailed ataxia clinical features.

The pooled figures below are the anchors — but read them through the lens of the whole module: each is a cohort-specific number, and the reasoning behind it is what transfers to the patient in front of you, not the digit itself.

Pooled Yields Across Pediatric NDD Cohorts

Notice that the all-epilepsy row (9 / 24 / 48%) recapitulates the broad-NDD row (10 / 36 / 41%) but with a wider WGS-over-WES gap — a reminder that the WES-vs-WGS verdict is phenotype-dependent, not universal. Where non-coding and structural pathology is enriched, the genome's extra reach starts to pay off; where it is not, the two converge.

Six take-home points

• CMA is not obsolete — it owns the dosage window (aneuploidy, large CNVs, UPD) that sequencing reads poorly, complementing WES at lower cost rather than competing with it
• Trio is essential for severe early-onset phenotypes — its ~2× yield is interpretive power (instant de novo calling, phasing), not deeper sequencing
• WGS > WES only in specific scenarios — post-WES-negative, leukodystrophies, atypical CP, because most interpretable disease variants are still coding
• Repeat expansions are a separate testing universe — Friedreich, SCA, CANVAS, DM1, HD, C9orf72 are invisible to short-read sequencing by the nature of the chemistry and need dedicated testing
• Yield is dynamic — reanalysis at 12–24 months yields new diagnoses in ~10–25% of unsolved cases as knowledge, not data, accrues
• Leukodystrophy is the special case — front-loaded MRI phenotyping pushes WES/WGS yield to 50–89%

Clinical utility ≠ diagnostic yield. A negative test can still inform recurrence counseling, and a positive one earns its value only if it changes something. The strongest evidence for that change sits in the NICU, where a diagnosis alters management in 38–50% of cases, and in IESS, where it enables precision therapy in 61.6% of genetically explained cases — because here the diagnosis maps onto a specific, time-sensitive intervention. Examples: KCNQ2 → carbamazepine; GLUT1 → ketogenic diet; SLC6A1 → valproate first-line; SCN1A → avoid sodium channel blockers. The decision to test, and which test, should ultimately be argued from this end — what action a result would change — not from the yield percentage alone.