Dataset: gnomAD constraint - an open, instant tolerability prior that cannot discriminate between our candidates
also called gnomAD access plan · LOEUF pull · human LoF constraint
gnomAD collects the genetic variation of hundreds of thousands of people and tells us which genes almost never tolerate being broken. It is free and instant. The awkward finding is that every single target we are considering is in that intolerant category, so the measure cannot help us choose between them.
The same thing, for a scientist
gnomAD supplies the organismal tolerability prior for each candidate through pLI and LOEUF, retrievable in one GraphQL call with no authentication. The census was run this cycle for all eight candidate genes and it produced a result that is itself the finding: **every candidate has pLI at or above 0.76 and seven of eight are essentially 1.0**, with LOEUF ranging from KMT2A's extreme 0.070 to AFF1's comparatively relaxed 0.502. Constraint therefore cannot rank this portfolio. Worse, if it were used as a gate it would have eliminated MEN1 (LOEUF 0.204), the one mechanism in this field that produced an approved paediatric drug. The correct phase-2 use of gnomAD is as a prompt to look for a developmental phenotype, not as a filter.
What question it answers, and what it does not
Answers: across roughly a million sequenced humans, does this gene tolerate a heterozygous loss-of-function variant? A LOEUF near 0.1 means almost no such variants are observed relative to expectation.
Does not answer: whether partial, reversible, time-limited pharmacological inhibition is tolerated. Those are different perturbations on different timescales in different tissues. MEN1 is the proof: LOEUF 0.204, and a menin inhibitor is approved and given to children.
How to reach it
One POST to https://gnomad.broadinstitute.org/api, no authentication 1:
{gene(gene_symbol:"MEN1", reference_genome:GRCh38){gene_id gnomad_constraint{pLI oe_lof oe_lof_upper mis_z exp_lof obs_lof}}}
oe_lof_upper is LOEUF, which is the field to use; pLI saturates at 1.0 and stops discriminating, which is exactly what happened across this portfolio.
The census, and the finding it produced
| Gene | pLI | LOEUF |
|---|---|---|
| KMT2A | 1.000 | 0.070 |
| MEN1 | 1.000 | 0.204 |
| KAT7 | 1.000 | 0.207 |
| DOT1L | 1.000 | 0.288 |
| MLLT1 (ENL) | 1.000 | 0.330 |
| CBX8 | 0.9999 | 0.363 |
| PSIP1 (LEDGF) | 1.000 | 0.373 |
| AFF1 | 0.765 | 0.502 |
Two things follow. First, pLI is useless here - seven of eight saturate. Second, the LOEUF ordering does carry a signal, and it points the same way the rest of this review does: KMT2A is the gene you must not touch systemically, and LEDGF is the least constrained of the serious candidates. But LOEUF differences of 0.2 between genes of different length are within the estimator's noise, so this ordering is a hint, not a gate.
The access barrier
None. Open GraphQL, no key, fast, safe to automate, and the underlying data is aggregate allele-frequency information that carries no individual-level access restriction.
| resource | gnomAD (Genome Aggregation Database) constraint metrics |
|---|---|
| question answered | Does the human population tolerate loss of this gene, as an organismal prior on target tolerability? |
| access barrier | None. Aggregate data, no agreement required. |
| finding | Constraint cannot rank this portfolio: seven of eight genes saturate pLI at 1.0. Used as a gate it would have eliminated MEN1, the only mechanism in this field with an approved paediatric drug. Correct use is as a prompt to search for a developmental phenotype, not as a filter. |
| status | run |
| compute | laptop; no GPU |
access
{
"auth": "none",
"endpoint": "https://gnomad.broadinstitute.org/api",
"method": "POST GraphQL",
"preferred_field": "oe_lof_upper (LOEUF); pLI saturates and does not discriminate"
}census this cycle — the numbers behind the chart
[
{
"gene": "KMT2A",
"pLI": 1,
"loeuf": 0.0704
},
{
"gene": "MEN1",
"pLI": 0.9999999992,
"loeuf": 0.2042
},
{
"gene": "KAT7",
"pLI": 1,
"loeuf": 0.2067
},
{
"gene": "DOT1L",
"pLI": 1,
"loeuf": 0.2879
},
{
"gene": "MLLT1",
"pLI": 0.9999984,
"loeuf": 0.3304
},
{
"gene": "CBX8",
"pLI": 0.9999132,
"loeuf": 0.3634
},
{
"gene": "PSIP1",
"pLI": 0.9999984,
"loeuf": 0.373
},
{
"gene": "AFF1",
"pLI": 0.7645,
"loeuf": 0.5017
}
]- Using constraint as a gate would have killed the menin programme. Any automated druggability scoring that includes LOEUF as a penalty term will make the same mistake, silently, across the whole portfolio.
- Constraint reflects germline heterozygous loss in a population that survived to be sequenced. It is systematically blind to effects that are lethal in utero and to effects that only appear under pharmacological, whole-body, sustained inhibition.
- Is there a constraint metric that distinguishes tolerance of partial inhibition from tolerance of complete loss, and does any published work calibrate LOEUF against clinical drug tolerability?
- Do any of these genes have recessive or hypomorphic human phenotypes that would be more informative than heterozygous LoF constraint?
- Is KMT2A's extreme LOEUF of 0.070 driven by Wiedemann-Steiner syndrome ascertainment, and does that change how it should be read?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
The gnomAD GraphQL API is open and key-free, and a constraint census run this cycle returned: KMT2A pLI 1.0 / LOEUF 0.070; MEN1 pLI 0.9999999992 / LOEUF 0.204; KAT7 pLI 1.0 / LOEUF 0.207; DOT1L pLI 1.0 / LOEUF 0.288; MLLT1 pLI 0.9999984 / LOEUF 0.330; CBX8 pLI 0.99991 / LOEUF 0.363; PSIP1 pLI 0.9999984 / LOEUF 0.373; AFF1 pLI 0.765 / LOEUF 0.502.
gnomAD GraphQL API, gnomad_constraint for eight candidate genes (GRCh38) · Broad Institute 2026What this evidence cannot show
Constraint estimates depend on gene length and on the reference cohort's ascertainment; differences of a few tenths of a LOEUF between genes of different length are not reliably interpretable. The gnomAD version served by the API was not recorded from the response.
Evidence that argues against this (1)
- MEN1, with LOEUF 0.204, is the target of an approved and tolerated drug. Constraint as a kill criterion would have removed the one mechanism that worked, which is the strongest available argument against using it as a gate. source 2026
constraint cannot rank this portfolio and would have disqualified menin
constraint cannot rank this portfolio and would have killed the menin programme; it must not be used as a gate
KMT2A LOEUF 0.070, the most constrained gene in the beat