Oncofusion

start / graph / strategy-open-ground-uncontested-questions

Hypothesismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q9-landscape

Open ground: the questions in this graph that nobody in the landscape appears to be working on

also called uncontested questions · where a small lab can win · white space analysis

We compared the 777 open questions recorded in this knowledge graph against what every major group in the field is actually publishing. Six areas came back with real questions and no visible competitor, and that is where a small lab should spend its time.

The same thing, for a scientist

A white-space analysis crossing the graph's accumulated open questions (777 items across 12 domain folders as of this cycle) against the published output of the groups catalogued in the landscape beat. The finding is not that the field is idle, but that its effort is extraordinarily concentrated: menin chemistry, menin resistance, adult NPM1-mutant clinical development, DOT1L, enhancer architecture, breakpoint cataloguing and cell of origin absorb almost all of it. Six clusters carry substantive open questions with no identifiable group working on them, and they share a property that matters for a small laboratory: each is answerable primarily by analysis of data that already exists, or by one cheap experiment, rather than by outspending a competitor.

Go deeper
The six areas of open ground2 deeper

1. Breakpoint position as a biological and prognostic variable. 37 of the graph's open questions touch breakpoints. Frankfurt holds nucleotide-resolution breakpoints for 3,401 patients but publishes them as a catalogue, not as a variable; Interfant holds breakpoint-derived MRD targets for 249 infants linked to protocol-quality outcome data but analysed MRD, not the breakpoint. No group has joined the two. This is the clearest open ground in the entire survey and it requires no laboratory.

2. The reciprocal fusion product. 14 open questions. Marschalek has argued for two decades that the reciprocal AF4-MLL product matters; almost no one tests it, and a Menendez-Marschalek collaboration produced a negative result on one proposed mechanism. Yet the graph contains a ready-made natural experiment: spliced KMT2A::MLLT10 patients express no reciprocal product and can be compared with reciprocal-expressing patients inside the existing DCAL cohort. Nobody has done it.

3. Condensates and subnuclear localisation of the fusion. 15 condensate and 4 nuclear-speckle open questions, and not one group in this landscape works on it. The graph records that nobody has published live-cell imaging plus FRAP on an endogenously tagged KMT2A::AFF1 in a t(4;11) line. That is a single, cheap, definitive experiment on a mechanism class the field increasingly invokes and has never tested for this fusion.

4. Infant-specific menin biology. 138 open questions mention infants; the clinical menin programme's centre of gravity is adult NPM1-mutant AML with median ages of 66-69. Every claim about menin inhibition in infants is an extrapolation across a fifty-year age gap, a different genotype and a different cell of origin. Nobody owns the infant question because the commercial incentive points elsewhere.

5. Cross-validation of non-menin dependencies. PSMB8, PTBP1/PKM2, ARID2, C/EBP, and others are each reported as a KMT2A-fusion dependency by exactly one group, in different models, with different readouts. Nobody has evaluated them against each other and against DepMap in a single consistent framework. That is pure analysis and it directly serves the lab's thesis that menin is not the only obligate dependency.

6. Non-coding and regulatory drivers. The graph notes that every published mutation-burden figure for infant KMT2A-r is a coding count, and that a whole-genome cohort has never been analysed for recurrent regulatory mutations at adequate power. If a missed co-driver exists in this 'silent genome' disease, that is where it is hiding, and no group in the landscape is looking.

Why these are open, which is the more useful question

Three structural reasons, and recognising them is how you predict the next piece of open ground rather than just finding this one.

Incentive mismatch. Menin inhibitors are commercially valuable in adult NPM1-mutant AML, which is roughly three times more common than KMT2A rearrangement in AML and has a large, tolerant, insured adult population. Infant ALL has a few hundred cases a year worldwide. Industry effort follows the adult indication; the infant questions are left standing.

Capability mismatch. The groups holding infant material (Interfant, Frankfurt, COG) are clinical and diagnostic operations that do not run mechanism programmes. The groups running mechanism programmes (Boston, Oxford, Barcelona) do not hold the large annotated clinical cohorts. Questions that need both sit in the gap, unowned. This is exactly the gap a computational lab with good collaboration instincts can occupy.

Method mismatch. Condensate biology needs live-cell imaging and FRAP; the KMT2A field is a ChIP-seq and mouse-genetics field. Whole-genome non-coding analysis needs statistical genomics at scale; the field counts coding mutations. Nobody is avoiding these questions on the merits - they are simply outside the methodological habits of the people who own the models.

Ranked, by value divided by cost
  1. Breakpoint-as-variable (Frankfurt + Interfant). Zero wet-lab cost, two collaborations, a twenty-year-old question, publishable alone. See hyp-breakpoint-position-prognostic-interfant.
  2. Non-menin dependency cross-validation. Zero cost, public data only, directly serves the lab's central thesis, produces a ranked target list for every subsequent project.
  3. Condensate/FRAP on endogenously tagged KMT2A::AFF1. One focused experiment, one cell line, definitive either way, and it settles a mechanism-class question the field keeps asserting.
  4. Fusion-driven surfaceome beyond PROM1. Public expression data; feeds directly into the CAR programmes at Oxford/Imperial and is useful to them.
  5. Reciprocal-product natural experiment in the DCAL cohort. Needs Frankfurt, but only their existing annotation.
  6. Non-coding driver analysis in an infant whole-genome cohort. Needs cohort access (St. Jude Cloud, TARGET or Interfant); highest potential payoff, highest access risk.
Facts and figures
statementThe Iqbal Lab's highest-return strategy is not to compete on menin biology but to occupy six specific questions that carry real open problems in this graph and have no identifiable group working on them: breakpoint position as a biological and prognostic variable; the reciprocal fusion product; condensates and subnuclear localisation of KMT2A::AFF1; infant-specific menin pharmacology and resistance; cross-validation of non-menin dependencies; and non-coding/regulatory drivers in the 'silent' infant genome.
originliterature-gap
prior probability0.6
predictions
[
  "A systematic literature search for each of the six clusters will return no primary paper by any group catalogued in the landscape beat within the last five years.",
  "Breakpoint position will show at least one statistically detectable association with outcome or with MRD kinetics once tested in the Interfant-06 cohort, because the Frankfurt age-dependent breakpoint shift implies breakpoint carries biological information.",
  "Cross-validating the single-group non-menin dependencies against DepMap will find that at least one is not reproducible across independent KMT2A-r cell lines.",
  "Endogenously tagged KMT2A::AFF1 will either form FRAP-mobile foci or not; either result is publishable because the field currently assumes the answer without data."
]
experiments
[]
analyses
[
  "ana-interfant-breakpoint-prognostic-reanalysis",
  "ana-non-menin-dependency-cross-validation"
]
falsifiers
[
  "Finding a recent primary paper from any landscape group directly addressing any of the six clusters falsifies that cluster's 'open' status.",
  "If breakpoint position shows no association with outcome, MRD kinetics, retained domain content or menin-inhibitor sensitivity in an adequately powered cohort, area 1 closes.",
  "If the Interfant breakpoint sequences were not archived, area 1 becomes inaccessible rather than merely open, which is a different and more terminal problem.",
  "If DepMap and independent cell lines reproduce all the single-group non-menin dependencies, area 5 is already effectively done and adds nothing."
]
open question counts by theme
{
  "total_open_questions_in_graph": 777,
  "menin": 158,
  "infant": 138,
  "lineage_switch": 61,
  "fetal": 50,
  "resistance": 41,
  "breakpoint": 37,
  "car_immunotherapy": 37,
  "surface_antigen": 31,
  "condensate_or_speckle": 19,
  "degradation": 18,
  "reciprocal_product": 14,
  "topoisomerase_or_rloop": 9,
  "noncoding_or_lncrna": 4,
  "method": "case-insensitive keyword counts over open_questions[] extracted from all node files in data/graph/nodes; counts overlap and are indicative, not a partition"
}
How this could bite us back
  • Uncontested ground is sometimes uncontested because it is unfundable, unpublishable or already known to be a dead end; the survey cannot distinguish 'neglected' from 'correctly abandoned'.
  • Two of the six areas depend entirely on collaborations (Frankfurt, Interfant) that may take longer to negotiate than a grant cycle.
  • If the lab occupies a niche and then succeeds, a better-resourced group can enter and overtake within a year, because the barrier to entry that kept them out was attention, not capability.
Still open
  1. Is 'nobody is working on it' true, or merely 'nobody has published on it'? Conference abstracts and preprints were not systematically searched in this pass and should be before any of these six areas is committed to.
  2. Which of the six areas would a reviewer consider the lab credible to attempt? Open ground the lab cannot plausibly occupy is not an opportunity.
  3. Does occupying an uncontested question mean it is undervalued, or that the field has already judged it uninteresting? The second possibility deserves an honest test for each area before committing.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The Frankfurt DCAL series holds nucleotide-resolution genomic breakpoints for 3,401 acute leukaemia patients analysed 2003-2022 and publishes them as a partner-gene catalogue with an age-dependent breakpoint shift, not as a prognostic variable linked to outcome.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, ... Marschalek R · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    A referral series without systematic outcome linkage. The claim that breakpoint position has never been analysed as a prognostic variable is an argument from a literature search, not a proof of absence; a negative or buried analysis may exist.

    Evidence that argues against this (1)
    • AREA 1 (breakpoint position as a biological and prognostic variable) is not open ground. The entry says 'No group has joined the two' and calls this 'the clearest open ground in the entire survey'. It was joined in 2013, and by the Frankfurt group itself. Emerenciano, Meyer, Mansur, Marschalek and Pombo-de-Oliveira analysed 545 childhood leukaemias aged 24 months or under (385 ALL, 160 AML), localised the genomic breakpoint in 30 KMT2A-rearranged cases, and reported 'first evidence that MLL-r patients with poor outcome preferentially displayed chromosomal breakpoints within MLL intron 11', concluding that 'MLL breakpoint localization is of clinical importance and should be considered as a novel outcome predictor for MLL-r patients'. Claus Meyer and Rolf Marschalek are the second and fourth authors, so the premise that Frankfurt 'publishes them as a catalogue, not as a variable' is false as a statement about the group, whatever is true of any single paper. What survives, and it is worth more than the original claim: the 2013 analysis had only 30 breakpoint-localised cases, no multivariable adjustment for age, white-cell count or MRD, and a single-country diagnostic referral series rather than a protocol cohort. So the honest description of area 1 is a specific, directional, under-powered published hypothesis awaiting confirmation in a trial cohort - which is a better project than an unexplored question, because it can be pre-specified, and a much worse pitch if presented to Interfant or Frankfurt as though nobody had thought of it. Read: abstract only. Emerenciano M et al. 2013
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Interfant-06 measured MRD in 249 infants with KMT2A-rearranged ALL using DNA-based PCR of rearranged KMT2A genes, which requires patient-specific genomic breakpoint knowledge, and analysed MRD rather than breakpoint characteristics as the prognostic variable.

    PMID 33405950DOI
    What this evidence cannot show

    That the breakpoint sequences were archived and remain retrievable is an inference from the assay methodology, not a stated fact.

    Evidence that argues against this (1)
    • AREA 2 (the reciprocal fusion product) is the case the brief warns about: uncontested because it was tried in the relevant system and failed. The entry says 'almost no one tests it'. Prieto, Marschalek, Kuhn, Bursen, Bueno and Menendez tested it directly and negatively. Expressing AF4-MLL in human cord-blood-derived CD34+ HSPCs gave enhanced long-term haematopoietic reconstitution in primary immunodeficient recipients but no self-renewal on serial transplantation, and expression of AF4-MLL in primary neonatal CD34+ HSPCs failed to produce any phenotypic or haematological sign of disease and was therefore not sufficient to initiate leukaemia within a 36-week follow-up. The authors attribute the discrepancy with the earlier murine result to species-specific genetic and epigenetic determinants. That is a published negative result in the human cell type this project cares about, from the two groups best placed to get a positive one, and it explains the silence far better than lack of interest does. A caveat that cuts the other way and should be recorded honestly: the same Barcelona-Frankfurt collaboration later showed that co-expression of AF4-MLL with MLL-AF4 in human embryonic stem cell haemato-endothelial differentiation does cooperate, promoting hemato-endothelial precursor emergence with H3K79me3 enrichment at HOX-A cluster genes (Bueno et al., Haematologica 2019, PMID 30679325, DOI 10.3324/haematol.2018.202044). So the live question is not 'does the reciprocal matter' - that has been asked twice - but 'does it matter as a co-factor rather than as an initiator', and the experiment that answers it is a two-fusion comparison, not a single-fusion one. Caveats: n is small in both studies, both use surrogate developmental systems rather than patient material, and 36 weeks may be short for a prenatal-origin disease. Read: abstracts only. Prieto C et al. 2017
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The clinical menin-inhibitor programme's registration and combination cohorts are adult and predominantly NPM1-mutant, with median ages of 69 (KOMET-001 phase 2, n=92) and 66 (KOMET-007 combination, n=67), leaving infant KMT2A-rearranged biology addressed only by extrapolation.

    Ziftomenib in Relapsed or Refractory NPM1-Mutated AML · Wang ES, et al. · Journal of Clinical Oncology 2025
    PMID 40997296DOI
    What this evidence cannot show

    Revumenib's first-in-human trial did include paediatric sites and a KMT2A-rearranged population, so 'adults only' would be too strong; the claim is about where the field's centre of gravity and combination development sit, not about total exclusion of children.

    Evidence that argues against this (1)
    • AREA 4 (infant-specific menin biology) is only half open, and the half that is closing is closing fast. The entry says 'Nobody owns the infant question because the commercial incentive points elsewhere.' Two registry records contradict the ownership claim, both retrieved 2026-09-18. COG AALL2121 (NCT05761171) is a phase 2 of revumenib plus chemotherapy in relapsed/refractory KMT2A-rearranged infant leukaemia, ages 1 month to under 6 years with diagnosis under 2 years, ACTIVE_NOT_RECRUITING at 78 patients across 84 sites, primary completion December 2027 - and its protocol-registered exploratory outcomes are paired pre/post RNA-seq of the KMT2A-R transcriptional programme and menin occupancy by CUT&RUN with CUT&Tag for histone marks, correlated with clinical response and pharmacokinetics. TINI 2 (NCT05848687) is a RECRUITING frontline total-therapy protocol in newly diagnosed infants (maximum age 365 days at diagnosis, n=90, 24 sites) whose stated secondary objective is to determine the minimum safe and biologically effective dose of ziftomenib in combination with chemotherapy. So the infant menin question is owned, by two academic sponsors, with the obvious correlative assays already written into one protocol. What remains genuinely open is narrower and should be stated that way: AALL2121's primary MRD endpoint is restricted to patients 6 months and older, with the under-6-month group relegated to a descriptive exploratory outcome; AALL2121 excludes patients previously treated with revumenib, so it generates no post-menin-inhibitor resistance material; and TINI 2 gives blinatumomab alongside ziftomenib, so its menin correlatives will be confounded. The open ground is the under-6-month band, the resistance cohort, and any assay the two protocols do not list - not 'infant menin biology' as a whole. Caveat: registry records describe planned analyses and eligibility, not enrolment achieved or specimen availability. Children's Oncology Group; Gruber TA (TINI 2 sponsor) 2026
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Non-menin KMT2A-fusion dependencies are being reported by single groups in mutually incompatible model systems: the immunoproteasome subunit PSMB8 (Greifswald), PTBP1/PKM2 metabolic dependency in MLL-ENL (Slany), and stage-dependent SWI/SNF subunit requirements in MLL-AF9 (Versiti), with no cross-validation in a common framework.

    PMID 38049829DOI
    What this evidence cannot show

    This citation evidences one of the three findings; the absence of cross-validation is an assessment from the survey rather than a cited result. Different fusion partners (MLL-ENL, MLL-AF9, KMT2A-complex generally) may genuinely have different dependencies, in which case 'inconsistency' is biology rather than a gap.

    Evidence that argues against this (1)
    • AREAS 3 and 5, and the same paper bears on both. Area 3 claims that condensates and subnuclear localisation of the fusion are worked on by 'not one group in this landscape'. True of the landscape as catalogued, but false as an implication that the mechanism class is untested in MLL-fusion leukaemia. In January 2026 Nature Communications published work showing that the RNA-binding protein AKAP95 is required for MLL-rearranged leukemogenesis; that its phase separation and RNA binding modulate Pol II recruitment into condensates genome-wide; that AKAP95 interacts with the MLL1 translocated fragment and that their partial co-condensation strengthens AKAP95 binding at MLL-AF9 target genes; that AKAP95 loss downregulates MLL-AF9 targets and impairs MLL-AF9-driven leukemogenesis while being dispensable for normal and regenerative haematopoiesis; and that a designed peptide (JD-PI95) bridging AKAP95 to HSP70 impairs its phase separation and inhibits MLLr leukaemia cell proliferation. So condensate biology in this disease is not merely invoked - it has been perturbed pharmacologically. Area 5 is affected by the same paper for a different reason: AKAP95 is precisely another single-group, single-model non-menin dependency claim, published after this entry's candidate list was drawn. A curation exercise over a snapshot list is chasing a moving target, and the analysis is worth doing only if it is set up to ingest new claims as they appear. One further caution on area 3 that needs no citation to see: the 'single, cheap, definitive experiment' proposed - live-cell imaging plus FRAP on an endogenously tagged KMT2A::AFF1 in a t(4;11) line - is not straightforward, because a t(4;11) cell carries a wild-type KMT2A allele and wild-type AFF1 alleles alongside the fusion, and any tag knocked into the KMT2A 5' end or the AFF1 3' end labels the wild-type protein as well as the fusion. Isolating fusion-specific signal requires tagging across the breakpoint junction or using dual orthogonal tags, which is a substantially harder build than the entry implies, and that difficulty is a better explanation for the silence than lack of interest. Caveats on the cited paper: MLL-AF9 and murine models throughout, not KMT2A::AFF1 and not infant B-ALL; the peptide is a tool compound, not a drug. Read: abstract only. Yan X et al. 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Relapse in KMT2A-rearranged leukaemia separates into at least two mechanistically distinct classes, with drug-response gene alterations and a chemotherapy-exposure mutational signature present in early relapse ALL (9-36 months) but essentially absent in very early relapse (<9 months), implying a pre-existing rather than acquired resistant population in the latter.

    The genomic landscape of relapsed infant and childhood KMT2A-rearranged acute leukemia · Ahlgren L, ... Hagstrom-Andersson AK · Nature Communications 2025
    PMID 41062506DOI
    What this evidence cannot show

    n=36 for discovery sequencing with small subgroup denominators. The absence of a coding driver in very-early relapse does not establish that a non-coding one exists; it is consistent with a non-genetic (epigenetic or cell-state) mechanism.

    Evidence that argues against this (1)
    • AREA 6 (non-coding and regulatory drivers). The entry says 'a whole-genome cohort has never been analysed for recurrent regulatory mutations at adequate power' and that 'no group in the landscape is looking'. The first half is defensible; the second half misstates the reason, and the real obstacle is not access. Andersson et al. performed whole-genome, exome, RNA and targeted DNA sequencing on 65 infants (47 MLL-rearranged, 18 non-MLL-rearranged) and 20 older children with MLL-rearranged leukaemia, and reported that infant MLL-R ALL has one of the lowest somatic mutation frequencies of any sequenced cancer, with the predominant leukaemic clone carrying a mean of 1.3 non-silent mutations. Two things follow. First, the whole-genome cohort exists, it is a St. Jude PCGP cohort, and it is therefore reachable through St. Jude Cloud under a data-access agreement rather than through a collaboration - so this is an open-data project, not a blocked one, which strengthens the case for doing it. Second, and against the entry: at a mean of 1.3 non-silent mutations per clone with n=47 whole genomes, the power to detect recurrent regulatory events is very low and a null result would be uninterpretable. That is a genuine technical barrier and it is the most likely reason the analysis has not been reported. The honest version of area 6 is that it needs a pooled cohort across PCGP, TARGET and any European whole-genome series before it is worth starting, not that it is cheap ground nobody noticed. Caveats: the paper's headline mutation counts are coding-focused, and its abstract does not state whether a non-coding recurrence analysis was attempted and found nothing - which is itself a question worth putting to the authors. Read: abstract only. Andersson AK et al. 2015
Connected to