Oncofusion

start / graph / concept-phase2-go-no-go-register

Explainermoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q14-phase2-readiness

The phase-2 go/no-go register: the numbers we agree to before we look

also called pre-registered decision gates · kill criteria · phase-2 stopping rules

Before starting any of the deeper drug-discovery work, we write down in advance the exact number that would mean keep going and the exact number that would mean stop. Writing them down first is the only way to stop ourselves explaining away a bad result later.

The same thing, for a scientist

A pre-registered decision framework for phase 2. Every candidate programme in this graph gets three gates: G1 feasibility (can the assay or dataset be built at all), G2 biology (does the intended mechanism produce the intended effect at the intended magnitude), G3 window (is there a therapeutic index against normal haematopoiesis). Each gate carries a GO threshold and a STOP threshold that are numeric, directional and specified before the experiment. Between the two thresholds is an explicitly declared ambiguous zone with a single named tie-breaker, so ambiguity cannot be used as a licence to continue indefinitely. The register exists because the commonest failure mode in academic target validation is not a wrong hypothesis, it is a hypothesis that was never given a chance to be wrong.

Go deeper
The three gates2 deeper

G1 - Feasibility. Can we actually get the data or run the assay, at this institution, this year, with these credentials? A dataset behind a data-access agreement we have not applied for is a G1 failure, not a G2 question. Most of the dataset entries in this beat exist to answer G1 honestly.

G2 - Biology. Does perturbing the target produce the intended effect on the fusion output programme, at a magnitude that matters? The standard readout for this graph is HOXA9/MEIS1 transcript suppression plus a differentiation-marker shift plus a growth effect in KMT2A::AFF1 lines, with a non-KMT2A-rearranged line as specificity control.

G3 - Window. Is the effect bigger in the leukaemia than in normal haematopoietic cells? The graph already holds the comparators that make this quantitative: DepMap lineage-controlled selectivity 1, normal-tissue expression from GTEx, and human loss-of-function constraint from gnomAD 2.

Why the ambiguous zone is declared in advance

A single GO threshold with nothing below it invites the result to be rounded up. Each gate therefore carries two numbers. Above GO, continue. Below STOP, terminate and write the negative entry into this graph. Between them, exactly one named tie-breaker runs and its result is binding. No candidate may sit in the ambiguous zone for two consecutive cycles.

Why potency is never the whole gate for menin

Menin uses one pocket for both KMT2A and JUND, with opposite transcriptional consequences 3. A gate written only on KMT2A displacement potency would score a compound that also strips JUND as a success. Every menin-directed gate therefore carries a selectivity term.

The register: one kill criterion per candidate

Each dossier carries its full gate table in its own data.go_no_go. The single most decisive STOP criterion per candidate is reproduced here so the whole portfolio can be read at a glance.

CandidateThe result that stops it
Menin MBM1 interfaceNot killable on target biology; it is an approved mechanism. The killable question is frontline infant value: if adding a menin inhibitor to Interfant-style induction does not raise the fraction of infants reaching consolidation in MRD-negative remission by at least 15 absolute points, the frontline programme stops.
LEDGF / PSIP1 IBDIf a tool compound or peptide that demonstrably displaces the KMT2A 140-160 peptide from the IBD (biophysical IC50 below 1 uM) fails to reduce HOXA9 by at least 50% in a KMT2A::AFF1 line at a non-cytotoxic concentration, the IBD is an interface that can be blocked without consequence, and the programme stops.
DOT1LEffectively already stopped as a single agent. Restarts only as a combination and stops permanently if DOT1L inhibition plus menin inhibition does not extend time to resistance by at least 2-fold over menin inhibition alone in a barcoded in vitro resistance panel.
ENL / MLLT1 YEATSIf ENL degradation in a KMT2A::AFF1 line does not suppress HOXA9 by at least 50%, or if the same degradation reduces healthy-donor CD34+ colony formation by 50% or more at the effective concentration, stop.
KAT7 / HBO1If catalytic-site inhibition at 10x the biochemical Ki does not reduce H3K14ac by at least 50% in KMT2A::AFF1 cells, or if the DepMap lineage-controlled selectivity (currently -0.242) does not replicate in an independent KMT2A-rearranged line panel, stop.
CBX8If the best available chemical matter (ChEMBL best Kd 120 nM, 15 potent rows) cannot be advanced to sub-100 nM cellular engagement within one medicinal-chemistry cycle, stop. The literature already contains a direct disagreement about whether CBX8 is required at all.
Fusion junction epitopeAlready failed G1 as a small-molecule target: no structure of an intact fusion protein exists and the junction region is predicted disordered. The only surviving version is a peptide-MHC or TCR-mimic programme, which stops if the junction peptide is not detected by immunopeptidomics on a KMT2A::AFF1 line.
CXXC-DNA interfaceIf a CXXC-binding molecule cannot be shown to spare the other CXXC-domain proteins in the proteome by at least 30-fold, stop. Selectivity, not potency, is the gate.
Degrader modalityIf a degrader does not achieve at least 90% target-protein loss for at least 48 h in primary infant KMT2A::AFF1 blasts ex vivo, stop. Degrader programmes fail on exposure in the relevant cell, not on chemistry.
Struck, redefined and demoted candidates (cycle 22) — what is no longer eligible to enter this register, and why2 deeper

A counter-evidence sweep of the interactome in cycle 21 argued that seven candidate targets should be struck, one redefined, one demoted and one reframed. Cycle 22 re-fetched every cited source, read each one, and recorded the audit in ana-target-list-rewrite-audit. Nine of ten decisions stand as decisions; four stand on a different reason than was given, and one — CCNT1 — survives only because a second reason was already there. Nothing below was ever a registered candidate with a gate table, so nothing is being removed from the table above; the point of recording it here is that the register is where a future cycle will look before proposing one of these again.

CandidateDecisionThe reason that carries itAudit verdict
PRMT1strikePrmt1 knockdown left transformation by most MLL fusions unaffected; only MLL-GAS7 and MOZ-TIF2 were dependent, and MLL-AF9 recruits KDM4C but not PRMT1 (PMID 26766589, read in full). Pan-essential at Chronos -1.399. First-in-human trial terminated (NCT03666988).Confirmed, reasoning replaced. PMID 31395602, cited to strike it, in fact concludes for PRMT1 inhibition in MLL-r ALL.
RBBP5strikeThe catalysis WRAD exists to enable — the MLL1 SET domain — is dispensable for MLL-AF9 leukemogenesis (PMID 24813891), and KMT2A fusions do not retain the SET domain.Confirmed as argued.
ASH2LstrikeSame, plus WRAD is not a fixed unit: without WDR5, MLL3 binds RAD and shows enhanced activity (PMID 27563068).Confirmed as argued.
CTR9strikeNo primary study tests CTR9 individually in KMT2A-fusion leukemia; MEN1 co-dependency r = 0.006 (p = 0.963); DepMap common-essential. Wilms predisposition (PMID 25099282) is a caution on top.Confirmed, reasoning demoted — see the menin consistency test below.
CDC73strikeDepMap common-essential at -1.122, 98.7% of lines depleted. HPT-JT predisposition (PMID 12434154) is a caution on top.Confirmed, reasoning demoted — same test.
CCNT1striker = 0.349 fails Holm in the 16-gene panel and falls below every survivor of the 2,743-gene deep-pool null; CDK9, not cyclin T1, carries the chemistry.Clinical leg fails. See below.
ELL2strikeNot the required SEC paralogue: Chronos -0.064 with 1.0% depleted, against ELL at -1.302 with 94.0%. Separately, ELL2 loss costs antibody secretion and the plasma-cell UPR (PMID 25238757).Confirmed, reasoning inverted: the cited paper shows ELL2 is not dispensable.
MEF2CredefinePhospho-dead knock-in mice have normal haematopoiesis yet resist fusion-driven leukemogenesis; MARK/SIK inhibition kills only MEF2C-phosphorylated cells (PMID 29431698).Confirmed. The handle is S222 phosphorylation and the upstream kinase, not MEF2C dosage — which also means the 5q14.3 haploinsufficiency objection does not apply to it.
MEIS1demoteAdult conditional knockout causes cell-autonomous HSPC loss and stem-cell exhaustion (PMID 24498346): the LSC function is the HSC function. Fails Holm in the deep pool at 0.067.Confirmed. MEIS1 stays as a readout — it is half the standard G2 endpoint below — and is not a target.
KDM6AreframeNot uniformly a tumour suppressor: sustains imatinib resistance in CML independently of demethylase activity (PMID 33456567); in AML high KDM6A tracks venetoclax tolerance (PMID 36720973).Confirmed. A therapy-selection variable in both directions, and a catalytic inhibitor would miss the CML mechanism.

Two entries replace the struck ones rather than leaving a gap, both created in cycle 22 after the KMT2A::AFF1 evidence a previous cycle could not find was located in full text: gene-kdm4c and gene-bod1l. Neither is a registered candidate. Both are pre-gate, and the gate each must clear first is named in its entry.

The menin consistency test, now a standing rule for this register

Three of the ten decisions invoked a safety argument of the form this gene is a germline tumour suppressor, therefore reducing its function in a child is unacceptable. Applied consistently that argument disqualifies menin.

MEN1 is a tumour-suppressor gene whose germline inactivation causes multiple endocrine neoplasia type 1, a syndrome whose defining lesion is neuroendocrine parathyroid disease, with the classic two-hit pattern in sporadic tumours. CDC73's syndrome, HPT-JT, is also a parathyroid tumour syndrome. And the predicted endocrine effect of menin inhibition is not hypothetical: revumenib's own label reports increased intact parathyroid hormone in 34% of 241 patients. The project pursues it anyway, correctly, because germline haploinsufficiency across a lifetime and partial reversible occupancy over months are different exposures.

The rule this register now carries: before a safety argument is used to stop a candidate, it is run against menin. If it would have killed menin, it does not kill anything here on its own. It may still appear in bite_back. It may not appear in a kill line. This is the same exposure-mismatch reasoning ana-target-safety-triage-v2 applies to loss-of-function constraint, extended to predisposition genetics.

What the CCNT1 finding does NOT mean for CDK9

The CCNT1 strike was argued partly on clinical mortality: P-TEFb inhibition was run in AML with 7 of 9 treatment-related deaths in the alvocidib arm. The arithmetic is correct and the inference is not. In that randomised phase II of three regimens in 90 relapsed/refractory AML patients, the alvocidib arm had the highest response rate of the three (28% vs 14% and 16%), was significantly older (median 62 against 55), received a pan-CDK inhibitor alongside cytarabine and mitoxantrone, and the authors concluded the regimen 'had an encouraging response rate and should be considered for further clinical development but should be used with caution in elderly patients' (PMID 30370956).

A second randomised trial of the same drug in 165 newly diagnosed AML patients found a significantly higher complete-remission rate than 7+3 without increased toxicity, and no survival difference (PMID 26022709). Same drug, opposite toxicity readings, and the real problem in the larger trial is that extra remissions did not become extra survivors.

So CCNT1 is struck — on its co-dependency, which genuinely fails — and the CDK9 programme in target-cdk9-inhibition-kmt2a-r is untouched by this.

A gap this exposes, recorded rather than fixed here because the table above belongs to the dossiers: CDK9 has a target entry in this graph and has never been a registered candidate with a gate table. It is the most mechanistically direct target in the fusion's machinery and the one with clinical-stage compounds, and it has no STOP criterion anywhere. Whoever owns the next phase-2 cycle should either register it or record why not.

Rules that make the criteria hard to argue around
  1. Every threshold has a unit and a direction. "Reduces HOXA9" is not a criterion; "reduces HOXA9 mRNA by at least 50% at 72 h at a concentration below the 72 h GI20" is.
  2. Every threshold has a named comparator. A selectivity number without the non-KMT2A-rearranged control line is uninterpretable.
  3. Effect sizes, not p-values. With three to six cell lines, a p-value mostly measures how many replicates were run.
  4. The negative result is a deliverable. A STOP outcome is written into this graph as its own entry with the data attached. A programme that ends without an entry has wasted the money twice.
  5. No re-specification after unblinding. If a threshold is changed, the change and its date go into the dossier provenance and the original threshold stays visible.
Facts and figures
standard G2 readoutHOXA9 and MEIS1 mRNA at 72 h by RT-qPCR, differentiation-marker shift by flow cytometry, and 7-day growth inhibition, in SEM and MV4-11 (KMT2A::AFF1), with a non-KMT2A-rearranged B-ALL line as specificity control.
standard G3 readoutColony-forming-unit assay on healthy-donor CD34+ cells at the concentration that achieves the G2 effect; GO requires less than 30% CFU loss.
ambiguous zone ruleExactly one named tie-breaker per gate, binding; no candidate may remain ambiguous for two consecutive cycles.
menin consistency testBefore a safety argument stops a candidate, run it against MEN1. MEN1 is strongly constrained, a germline tumour suppressor causing a parathyroid syndrome, and its approved inhibitor measurably raises PTH in 34% of patients. Any rule that would have killed menin does not kill anything here on its own; it belongs in bite_back, not in a kill line.
strike recording ruleEach struck candidate records its load-bearing reason separately from its supporting reasons and names the result that would reopen it, so that a load-bearing reason which later fails reopens the decision instead of being silently propped up by the others.
unregistered gapCDK9 (target-cdk9-inhibition-kmt2a-r) is the most mechanistically direct target in the fusion's machinery, has clinical-stage compounds, and has never been a registered candidate with a gate table. It has no STOP criterion anywhere in this graph.
gate definitions
{
  "G1_feasibility": "Data or assay obtainable at this institution within the cycle, with any access agreement already granted.",
  "G2_biology": "Intended mechanism produces intended effect at intended magnitude, with a specificity control.",
  "G3_window": "Effect larger in KMT2A-rearranged leukaemia than in normal haematopoietic cells, quantified against DepMap lineage-controlled selectivity, GTEx normal-tissue expression and gnomAD constraint."
}
candidates registered
[
  "target-phase2-menin-mbm1-dossier",
  "target-phase2-psip1-ibd-dossier",
  "target-phase2-dot1l-dossier",
  "target-phase2-enl-yeats-dossier",
  "target-phase2-kat7-dossier",
  "target-phase2-cbx8-dossier",
  "target-phase2-fusion-junction-dossier",
  "target-phase2-cxxc-dna-dossier",
  "target-phase2-degrader-modality-dossier"
]
struck cycle22
[
  {
    "gene": "PRMT1",
    "node": "gene-prmt1",
    "load_bearing_reason": "not a general MLL-fusion dependency (PMID 26766589); pan-essential; NCT03666988 terminated",
    "reason_withdrawn": "PMID 31395602 argues FOR PRMT1 inhibition in MLL-r ALL and cannot be cited to strike it",
    "reopens_if": "a KMT2A::AFF1-specific PRMT1 requirement is shown, or the FLT3-methylation combination is pursued as a separate programme"
  },
  {
    "gene": "RBBP5",
    "node": "gene-rbbp5",
    "load_bearing_reason": "rationale was enabling MLL1 SET catalysis, which is dispensable (PMID 24813891); fusions do not retain the SET domain",
    "reason_withdrawn": null,
    "reopens_if": "a non-catalytic RBBP5 requirement is demonstrated in a KMT2A::AFF1 line"
  },
  {
    "gene": "ASH2L",
    "node": "gene-ash2l",
    "load_bearing_reason": "same, plus WRAD is not a fixed unit (PMID 27563068)",
    "reason_withdrawn": null,
    "reopens_if": "a non-catalytic ASH2L requirement is demonstrated in a KMT2A::AFF1 line"
  },
  {
    "gene": "CTR9",
    "node": "gene-ctr9",
    "load_bearing_reason": "never tested individually in KMT2A-fusion leukemia; MEN1 co-dependency r = 0.006; DepMap common-essential",
    "reason_withdrawn": "germline-tumour-suppressor status alone, which fails the menin consistency test",
    "reopens_if": "CTR9 is tested individually in a KMT2A::AFF1 line with a fusion-output readout"
  },
  {
    "gene": "CDC73",
    "node": "gene-cdc73",
    "load_bearing_reason": "DepMap common-essential, Chronos -1.122, 98.7% of lines depleted",
    "reason_withdrawn": "HPT-JT germline predisposition alone, which fails the menin consistency test",
    "reopens_if": "a separation-of-function surface is found that spares the pan-essential role"
  },
  {
    "gene": "CCNT1",
    "node": "gene-ccnt1",
    "load_bearing_reason": "r = 0.349 fails Holm in the 16-gene panel and falls below every deep-pool survivor; CDK9 carries the chemistry",
    "reason_withdrawn": "the alvocidib treatment-related-mortality argument (PMID 30370956), which does not survive reading the trial",
    "reopens_if": "a cyclin-T1-selective mechanism distinct from CDK9 blockade is proposed"
  },
  {
    "gene": "ELL2",
    "node": "gene-ell2",
    "load_bearing_reason": "not the required SEC paralogue: Chronos -0.064 / 1.0% against ELL at -1.302 / 94.0%",
    "reason_withdrawn": "the word 'dispensable' — PMID 25238757 shows ELL2 loss costs antibody secretion and the plasma-cell UPR",
    "reopens_if": "an ELL2-specific requirement in KMT2A-fusion leukemia is reported"
  }
]
redefined cycle22
[
  {
    "gene": "MEF2C",
    "node": "gene-mef2c",
    "from": "MEF2C dosage / degradation",
    "to": "MEF2C S222 phosphorylation and the upstream MARK/SIK kinases",
    "basis": "PMID 29431698",
    "note": "the redefinition dissolves the 5q14.3 haploinsufficiency objection, because phospho-blockade does not lower protein level",
    "reopens_if": "Evidence that MARK/SIK inhibition FAILS to phenocopy MEF2C loss in KMT2A-rearranged cells. This is a redefinition rather than a strike: the S222-phosphorylation framing is what dissolved the 5q14.3 haploinsufficiency objection, because phospho-blockade does not lower protein level. If the kinase route does not reproduce the phenotype, the redefinition is wrong and MEF2C dosage returns as the open question it was.",
    "reopens_if_provenance": "PROPOSED BY THE COORDINATOR, cycle 35 (2026-09-19), derived from this row's own recorded `basis`. These three decisions were being watched with no reopening condition at all, so no paper could ever satisfy them and the watch script suppressed their hits as unactionable. Open to override by the PI."
  }
]
demoted cycle22
[
  {
    "gene": "MEIS1",
    "node": "gene-meis1",
    "from": "candidate target",
    "to": "fusion-output readout and biomarker",
    "basis": "PMID 24498346; fails Holm in the deep-pool null at 0.067",
    "note": "MEIS1 remains half of this register's standard G2 endpoint, which is the role it should have had all along",
    "reopens_if": "A demonstration that lowering MEIS1 PROTEIN is selectively lethal in KMT2A-rearranged cells, independently of HOXA9 programme readout. The demotion rests on a CRISPR dependency result (fails Holm in the deep-pool null at p=0.067), and a gene-knockout screen is the wrong instrument for a factor whose abundance is set post-translationally - USP22 protects MEIS1 from proteasomal degradation and its loss degrades MEIS1 and impairs KMT2A-r growth (PMID 42418681). Reopen if a protein-level intervention separates MEIS1 from its role as a programme readout.",
    "reopens_if_provenance": "PROPOSED BY THE COORDINATOR, cycle 35 (2026-09-19), derived from this row's own recorded `basis`. These three decisions were being watched with no reopening condition at all, so no paper could ever satisfy them and the watch script suppressed their hits as unactionable. Open to override by the PI."
  },
  {
    "gene": "KDM6A",
    "node": "gene-kdm6a",
    "from": "tumour suppressor, points the other way",
    "to": "context-dependent therapy-selection variable",
    "basis": "PMID 33456567, PMID 36720973",
    "note": "cuts both ways: high KDM6A tracks venetoclax tolerance, and a catalytic inhibitor would miss the demethylase-independent CML mechanism",
    "reopens_if": "Either (a) a non-catalytic modality - degrader or interface disruption - showing a KMT2A-rearranged-selective effect, since the recorded basis is explicitly that a catalytic inhibitor would miss the demethylase-independent mechanism; or (b) the DIRECTION of effect resolved within KMT2A-rearranged disease specifically, rather than inferred across contexts that point opposite ways. The demotion was made because it cuts both ways, so what reopens it is the ambiguity being removed, not a new association.",
    "reopens_if_provenance": "PROPOSED BY THE COORDINATOR, cycle 35 (2026-09-19), derived from this row's own recorded `basis`. These three decisions were being watched with no reopening condition at all, so no paper could ever satisfy them and the watch script suppressed their hits as unactionable. Open to override by the PI."
  }
]
added cycle22 pre gate
[
  {
    "gene": "KDM4C",
    "node": "gene-kdm4c",
    "kmt2a_aff1_evidence": "SD70 preferentially suppressed SEM, a t(4;11) KMT2A::AFF1 line, and the recruitment domain maps to the MLL 5' moiety retained in every fusion (PMID 26766589, full text)",
    "gate_it_must_clear_first": "G2 specificity: does SD70 sensitivity in SEM survive rescue by KDM4A or KDM4B re-expression? DepMap shows no KDM4C dependency in any KMT2A::AFF1 line, so the family-versus-gene question precedes everything else."
  },
  {
    "gene": "BOD1L",
    "node": "gene-bod1l",
    "kmt2a_aff1_evidence": "CRISPR tiling screen in iCas9-MV4-11, a KMT2A::AFF1 line, shows a strong dependency confined to the N-terminal Shg1 domain (PMID 38989615, full text)",
    "gate_it_must_clear_first": "G1 feasibility: no structure, no ligand, no probe, and a 3,051-residue protein with one small ordered functional domain. Before any chemistry, repeat the tiling result in FLT3-wild-type KMT2A::AFF1 lines (SEM, RS4;11) with a HOXA9 readout."
  }
]
reopening condition audit cycle35
{
  "problem": "MEF2C, MEIS1 and KDM6A carried reopens_if = null while remaining on the watch. A decision with no reopening condition cannot be reopened by any result, so it is permanent by construction - the exact failure the cycle-23 audit named.",
  "action": "Conditions proposed by the coordinator from each row's recorded basis; 14 previously suppressed hits become testable against them.",
  "still_open": "CDK9 has no STOP criterion. It is discussed in this register but is not a row in candidates_registered, struck, redefined or demoted, so there is nothing to attach a criterion to. Registering it is a scope decision, not a wording one."
}
How this could bite us back
  • Pre-registration can entrench a wrong threshold: if the G2 magnitude is set too high, a real but partial mechanism is discarded. The mitigation is that the register records the reasoning behind each number, so a threshold can be challenged on its logic before the data arrive rather than after.
  • Thresholds set on cell lines can be systematically wrong for primary infant blasts, which are less proliferative and harder to culture; a compound that fails G2 in SEM may still act on the patient's disease.
  • A register creates an incentive to choose easily measured endpoints over important ones. HOXA9 mRNA is measurable; durable remission in an infant is what matters.
  • A strike is an intervention on the project and its cost is asymmetric and invisible: a wrongly struck target is one nobody looks at again, and there is no experiment that ever reports the error. The cycle-22 audit found four of ten decisions resting on a reason that fails on reading, and in all four a second reason happened to hold. That the list was robust to its own reasoning being wrong means the reasoning was not doing the work, and a genuinely wrong decision would have looked the same.
  • The menin consistency test is a loosening rule, and loosening rules have their own failure mode. If germline tumour-suppressor status is a weak objection for CTR9 and CDC73, it is weak for menin too — which means the endocrine late-effects risk this graph already tracks for revumenib is being underweighted rather than the CTR9 risk overweighted. The test says apply one standard, not apply the lenient one.
Still open
  1. Who holds the authority to overrule a STOP criterion, and is that person different from the person whose programme it is?
  2. Should the G3 CFU threshold be set on healthy-donor adult CD34+ cells when the patients are infants, given that fetal and infant progenitors are the relevant normal comparator and are far harder to obtain?
  3. Is 50% HOXA9 suppression defensible as the G2 magnitude, or is the relevant quantity the fraction of fusion-bound loci that lose expression rather than the mean across a handful of genes?
Every source behind this entry (8)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    DepMap CRISPR gene-effect data at per-cell-line resolution is reachable programmatically through the Open Targets Platform GraphQL API, which is what makes numeric G3 thresholds enforceable in this project rather than aspirational.

    What this evidence cannot show

    The API response does not state which DepMap release the values come from, so the release version is unverified; any threshold quoted against a Chronos value must be re-derived if the release changes.

    Evidence that argues against this (1)
    • Programmatic reachability is an engineering fact and the register treats it as a decision input. What matters for a go/no-go is whether the DATA answer the question, and this project has since shown that per-line DepMap in this disease carries a lineage confound that no amount of API access fixes. Oncofusion knowledge graph (internal analysis) 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Human loss-of-function constraint for every candidate in this register is directly retrievable from the gnomAD GraphQL API, giving an organism-level tolerability prior that can be written into a gate before any experiment is run.

    What this evidence cannot show

    Constraint measures intolerance to lifelong heterozygous loss in a population and is a weak proxy for tolerability of partial, reversible pharmacological inhibition in a child over months. Every candidate here has pLI near 1, so constraint alone cannot discriminate between them.

    Evidence that argues against this (1)
    • MEN1 itself has pLI 0.9999999992 and LOEUF 0.204 in gnomAD, yet menin inhibitors are approved and tolerated in humans. High constraint therefore does not forbid pharmacological inhibition, and a gate built on constraint alone would have killed the one mechanism that worked. source 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Menin binds KMT2A and JUND in the same pocket with opposite transcriptional consequences, which is the structural reason a menin-directed gate must contain a selectivity term and not only a potency term.

    PMID 22327296DOI
    What this evidence cannot show

    Structural and cell-based; does not establish that JUND liberation is clinically consequential at approved menin-inhibitor exposures.

    Evidence that argues against this (1)
    • The shared-pocket structure gives the register its selectivity term, and the pocket fact alone does not quantify it: no affinity, no stoichiometry and no cellular occupancy ratio is available for MLL1 versus JUND, so the selectivity term cannot currently be given a value. Nature 2012
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    PRMT1 is not a general MLL-fusion dependency: a screen of Prmt1 shRNA across more than ten MLL and non-MLL oncogenic transcription factors found transformation by most MLL fusions unaffected, with only MLL-GAS7 and MOZ-TIF2 highly Prmt1-dependent, and MLL-AF9 recruiting KDM4C but not PRMT1.

    PMID 26766589DOI
    What this evidence cannot show

    Full text read in cycle 22, not abstract only. Mouse c-kit+ HSPC retroviral transformation assay; the fusions screened are not individually enumerated in the text, so the absence of KMT2A::AFF1 from the dependent set is inferred from the two fusions the authors name as dependent. A negative in a mouse colony assay is weaker than a negative in the human disease, and none exists.

    Evidence that argues against this (1)
    • The paper originally cited to strike PRMT1 concludes the opposite: it reports that MS023 killing parallels baseline FLT3 R972/973 methylation in MLL-r ALL and that MS023 plus PKC412 outperformed PKC412 alone in patient-derived xenografts, ending that 'abolishing FLT3 arginine methylation through PRMT1 inhibition represents a promising strategy to target MLL-r ALL cells'. A biomarker-selected PRMT1-inhibitor combination in this disease remains published and unrefuted; this register declines it rather than disproving it. Blood 2019
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The MLL1 SET domain, and thus its H3K4 methyltransferase activity, is dispensable for maintaining haematopoietic stem cells and for supporting MLL-AF9-driven leukemogenesis; what is lost on Mll1 deletion is H4K16 acetylation, and SIRT1 inhibition restores it.

    The histone methyltransferase activity of MLL1 is dispensable for hematopoiesis and leukemogenesis · Mishra BP, Zaffuto KM, Artinger EL, et al. · Cell Reports 2014
    PMID 24813891DOI
    What this evidence cannot show

    Abstract read. Mouse, MLL-AF9 rather than KMT2A::AFF1, and about wild-type MLL1's SET domain, which KMT2A fusions do not retain in any case. The study does not test RBBP5 or ASH2L: the strike on both is an inference from the dispensability of the catalysis they support, not a result.

    Evidence that argues against this (1)
    • SET-domain dispensability is a mouse haematopoiesis and leukemogenesis result, and the register uses it to deprioritise catalysis. That is probably right and it is also the same paper this graph records as contested — a later study reports endogenous MLL1 dispensable with MLL2 sustaining the disease, which would change which enzyme the register should be scoring. Cell Reports 2014
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The alvocidib trial used to argue that P-TEFb inhibition in AML produces toxicity rather than selectivity does not support that reading: the alvocidib-containing arm had the highest response rate of three (28% vs 14% and 16%), was significantly older (median 62 vs 55), used a pan-CDK inhibitor with cytarabine and mitoxantrone, and the authors recommended the regimen for further clinical development with caution in the elderly.

    PMID 30370956DOI
    What this evidence cannot show

    Abstract read. Arms of 35, 36 and 19 under a Simon two-stage design per arm, so neither the response advantage nor the 7-of-9 treatment-related-death imbalance is a powered estimate; the age imbalance is stated but not adjusted for in the abstract. The 7 of 9 figure is accurate — it is the inference from it that fails.

    Evidence that argues against this (1)
    • A second randomised trial of the same drug reports the opposite toxicity finding: in 165 newly diagnosed intermediate/adverse-risk AML patients, FLAM gave a significantly higher complete remission rate than 7+3 (70% vs 46%, p=0.003) without increased toxicity, with no overall or event-free survival difference. The durable objection to this class is that extra remissions did not become extra survivors, not treatment-related mortality. Zeidner JF et al. 2015
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.review

    MEN1 is itself a tumour-suppressor gene whose germline inactivation causes multiple endocrine neoplasia type 1, a syndrome defined by neuroendocrine parathyroid, pancreas, duodenum and pituitary lesions, which is why a rule disqualifying germline tumour suppressors would disqualify this project's own approved lead target.

    Multiple endocrine neoplasia type 1 and 2: from morphology to molecular pathology 1997. · Komminoth P · Verhandlungen der Deutschen Gesellschaft fur Pathologie 1997
    PMID 9474863
    What this evidence cannot show

    Abstract read. A 1997 review, used as a map of an uncontroversial clinical-genetics fact rather than as primary evidence. Nearly thirty years old; take any penetrance figure from a modern source. The two-hit mechanism is separately evidenced in ana-target-list-rewrite-audit E8 (PMID 9361035).

    Evidence that argues against this (1)
    • MEN1 being a tumour suppressor is the register's central safety term and the citation is a 1997 morphology-to-molecular-pathology review. The substantive version of this concern is quantitative — germline penetrance curves over decades versus weeks of reversible occupancy — and this graph's endocrine beat has since established that the two are not on one axis. Verhandlungen der Deutschen Gesellschaft fur Pathologie 1997
  8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The actionable variable for MEF2C is its S222 phosphorylation and the upstream MARK/SIK kinases, not MEF2C dosage: phospho-blocking knock-in mice have normal haematopoiesis yet resist fusion-driven leukemogenesis, and MRT199665 caused apoptosis and chemosensitivity only in MEF2C-phosphorylated cells and primary specimens.

    MEF2C Phosphorylation Is Required for Chemotherapy Resistance in Acute Myeloid Leukemia · Brown FC, Still E, Koche RP, et al. · Cancer Discovery 2018
    PMID 29431698DOI
    What this evidence cannot show

    Abstract read. AML rather than infant KMT2A::AFF1 B-ALL, and the PubMed abstract is garbled at the point where the driving fusion is named, so the driver in the knock-in experiment is taken as MLL-AF9 on the coordinator's pre-verified reading rather than confirmed from the abstract. MRT199665 is a tool compound with no clinical development, so the redefined target currently has no clinical-stage chemistry at all.

    Evidence that argues against this (1)
    • Making S222 phosphorylation the actionable variable moves the target from a transcription factor to the upstream kinases, which is a different and more tractable proposition. The register should say that explicitly, because it changes MEF2C from a no-go on undruggability to a go on a kinase. Cancer Discovery 2018
Connected to