Oncofusion

start / graph / ana-pentimento-methylation-ground-destination

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q12-analogy

Analysis (runnable): build a methylome cell-of-origin score for infant KMT2A-rearranged ALL and test it as a predictor of lineage destination

Take the chemical marks on the DNA of these leukemias at diagnosis, work out which normal blood cell each one came from, and see whether that tells you where the cell goes when treatment releases it.

The same thing, for a scientist

Adapt methylome-based cell-of-origin modelling, developed for CLL against a normal B-cell differentiation reference [E2], to infant KMT2A-rearranged ALL against a fetal and neonatal haematopoietic reference. Stage 1 is a feasibility assessment of whether an adequate reference exists. Stage 2 fits the model to published infant methylome cohorts [E1] and produces a per-case origin score. Stage 3 tests that score as a predictor of two outcomes: post-menin-inhibition lineage destination in the samples generated by `exp-menin-inhibitor-lineage-destination`, and retrospective lineage switch after CD19-directed therapy in banked diagnostic samples. Stage 1 may terminate the analysis, and saying so in advance is the point.

Go deeper
Stage 1 — the feasibility gate

Methyl-COOM works because normal B-cell differentiation has a dense, well-characterised methylation reference. Infant KMT2A-rearranged ALL arises prenatally from fetal liver or early bone marrow progenitors, and the equivalent fetal haematopoietic methylation reference is far thinner.

Stage 1 inventories what exists: public fetal liver and fetal bone marrow HSPC, LMPP, pre-pro-B and pro-B methylation data at array or WGBS resolution, and assesses whether the cell types spanning the plausible origin range are covered densely enough to fit a continuum model. If they are not, the correct output is a statement of what reference would need to be generated and by whom — which is a concrete collaboration ask, and a more useful deliverable than a badly fitted model.

This gate is written first because the temptation to proceed with an inadequate reference and report a score with false precision is the main way this analysis goes wrong.

Stages 2 and 3

Stage 2. Fit the decomposition on published infant methylome cohorts, separating the inherited component from the leukemia-acquired component. Output a per-case origin position with an uncertainty interval, and check face validity against the transcriptome-based cell-of-origin assignments already published for infant B-ALL 3.

Stage 3. Two predictor tests, both head-to-head against a transcriptomic comparator so that the claim is about which measurement is better rather than about whether methylation works at all.

  • Destination under menin inhibition, using material from exp-menin-inhibitor-lineage-destination.
  • Retrospective lineage switch after CD19-directed therapy, using banked diagnostic samples from switch cases and matched non-switch controls, with the strong caveat that switch is rare and any such series will be small and selected.

Add CHD4 status and chromatin-modifier alterations as covariates throughout, since the competing account attributes destination to those 4. The output should apportion between ground and acquired lesion rather than declaring one the winner.

Statistical discipline

Pre-specify the origin-score construction before it touches any outcome. Report area under the curve with confidence intervals for the methylome score and the transcriptomic comparator side by side, on the same cases. Lineage switch is rare, so the switch arm will be badly underpowered; state the achievable power in advance and report the analysis as exploratory rather than pretending otherwise. Cross-cohort batch effects between methylation platforms are severe and must be handled by within-cohort scoring, as the existing ana-lineage-switch-signature-scoring node already specifies for its own signatures.

Facts and figures
scriptanalyses/pentimento-methylation-ground/run.py
statusplanned
result summaryNot yet run. No result should be quoted from this node.
cost tierlow
computelaptop to small-cluster; CPU only
timelineStage 1 two weeks; Stages 2-3 three to four months, gated on Stage 1 and on destination data becoming available
stopping ruleIf Stage 1 finds no adequate fetal reference, report that and the generation requirement, and do not fit a score against an inadequate reference.
statistical planScore construction pre-specified before contact with outcomes; within-cohort scoring to avoid cross-platform batch effects; head-to-head comparator on the same cases; CHD4 and chromatin-modifier status as covariates; the lineage-switch arm declared exploratory with power stated in advance.
inputs
[
  "Infant MLL-rearranged and non-MLL-rearranged B-ALL methylome and transcriptome data, PMID 33983906",
  "Methyl-COOM model specification, PMID 32188505",
  "Public fetal liver and fetal bone marrow HSPC, LMPP, pre-pro-B and pro-B methylation references (inventory is Stage 1)",
  "Transcriptome-based cell-of-origin assignments for infant B-ALL, PMID 30679323",
  "CHD4 and chromatin-modifier status where available, PMID 35839448",
  "Destination outcomes from exp-menin-inhibitor-lineage-destination, and banked diagnostic samples from lineage-switch cases"
]
outputs
[
  "Stage 1 feasibility report: whether an adequate fetal haematopoietic methylation reference exists, and if not, precisely what would need generating",
  "per-case methylome-derived cell-of-origin position with uncertainty",
  "head-to-head area under the curve for the methylome score versus a transcriptomic comparator in predicting destination and switch",
  "apportionment of destination variance between inherited ground and acquired chromatin-modifier lesion"
]
tests hypotheses
[
  "hyp-pentimento-ground-predicts-post-drug-destination"
]
How this could bite us back
  • Fitting a continuum model against a sparse reference will produce origin scores that look precise and are not, and those scores could then be used to stratify children. The Stage 1 gate is the only thing preventing that and must not be skipped.
Still open
  1. Who holds fetal haematopoietic tissue and would be willing to generate the methylation reference if it does not exist?
  2. How many banked diagnostic samples from confirmed lineage-switch cases exist worldwide, and is that number large enough for the switch arm to be worth attempting at all?
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Genome-wide methylome and transcriptome data exist for 69 uniformly treated infants with de novo MLL-rearranged and non-MLL-rearranged B-ALL, with methylome signatures mapping onto chromatin states related to enhancer and transcriptional control in normal haematopoietic cells.

    PMID 33983906DOI
    What this evidence cannot show

    n=69 total across subtypes, so the KMT2A-rearranged subgroup is smaller still; adequate for model fitting, thin for predictor validation.

    Evidence that argues against this (1)
    • The analysis assumes the diagnostic methylome is a stable record that remains valid at the time of the outcome it predicts. Paired diagnosis-remission-relapse profiling in paediatric T-ALL, including DNA methylation arrays, found relapse-specific promoter hypomethylation together with acquired mutations in epigenetic modulators (SUZ12, WHSC1, SMARCA4). If the methylome moves under therapy, a diagnostic cell-of-origin score is predicting a destination in a landscape that has since changed. Where paired diagnosis-relapse methylomes exist, the analysis should quantify drift and report the predictive performance conditional on it, rather than treating the diagnostic score as a fixed covariate. Kunz JB et al. 2015
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Methylome-based cell-of-origin modelling separates the epigenetic pattern inherited from the tumour-initiating cell from tumour-acquired changes using a linear regression model against a normal differentiation reference, and is proposed as generalisable to other malignancies.

    PMID 32188505DOI
    What this evidence cannot show

    The method's performance depends entirely on reference density, which is the subject of the Stage 1 gate.

    Evidence that argues against this (1)
    • Methyl-COOM was developed and validated in CLL, an indolent mature B-cell malignancy with a well-mapped normal B-cell differentiation continuum to regress against, and its decomposition into inherited plus tumour-acquired components is an assumption of the model rather than a measured separation. No published application to an acute leukaemia was found in this search. The source-domain analogue has the same difficulty: in conservation, X-ray imaging of pentimenti superimposes features from all paint layers and intensity depends jointly on layer thickness and pigment opacity, so the earlier layer is recovered by guided inference, not read off. The analysis should therefore include an identifiability check — for example, whether the fitted inherited component is stable to the choice of reference continuum — before any predictive claim is made. Verification note: the Gooch and Tumblin record was confirmed from the publisher's listing and abstract; full text was paywalled. Gooch AA et al. 2007
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Genome-wide analysis of infant B-ALL established the cellular origin of the disease and identified clinical prognostic markers including reciprocal AFF1::KMT2A expression, providing a transcriptome-based origin assignment against which a methylome-derived score can be checked for face validity.

    PMID 30679323DOI
    What this evidence cannot show

    Provides a comparator assignment, not a gold standard; agreement between the two would be reassuring and disagreement would be uninterpretable without an independent truth.

    Evidence that argues against this (1)
    • Searched for a published head-to-head in which a diagnostic methylation cell-of-origin score outperformed transcriptomic or genetic markers for predicting outcome or lineage destination in KMT2A-rearranged leukaemia. None found. Note the direction of the one comparison available inside this entry's own evidence: the prognostic discriminator identified in infant B-ALL was expression of the reciprocal AFF1::KMT2A fusion, a transcriptional feature. The analysis should therefore include the reciprocal-fusion status and a transcriptomic baseline as explicit comparators, not as covariates to adjust away. Recorded so the gap is known to have been checked. Q12-analogy-counter (search record) 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Lineage switching in MLL/AF4 leukemia is recurrently associated with altered expression, splicing or mutation of chromatin modifiers notably CHD4, whose perturbation alone induces myeloid gene expression, making chromatin-modifier status a mandatory covariate in any destination model.

    Epigenetic regulator genes direct lineage switching in MLL/AF4 leukemia · Tirtakusuma R, Szoltysek K, Milne P, et al. · Blood 2022
    PMID 35839448DOI
    What this evidence cannot show

    Establishes the competing mechanism; the analysis is designed to apportion between the two rather than to defeat it.

    Evidence that argues against this (1)
    • Association between lineage switching and chromatin-modifier alteration is drawn from a small number of switched cases, and CHD4 perturbation inducing myeloid gene expression is a sufficiency result in a model system. Neither establishes that the modifier changes precede the switch in patients rather than following it. Oncofusion knowledge graph (internal analysis) 2026
Connected to