Oncofusion

start / graph / ana-phase2-dataset-depmap

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q14-phase2-readiness

Dataset: DepMap CRISPR essentiality - the bulk download is now behind a captcha, and this changes how phase 2 must be built

also called DepMap access plan · Chronos data pull · Open Targets DepMap route

DepMap tells us which genes cancer cells cannot live without, and it is the backbone of every target decision in this project. As of now you can no longer download the big files with a script - a robot check blocks it - so we have to go through a different door or use an older copy.

The same thing, for a scientist

DepMap answers the single most important phase-2 question for every candidate: is the target more required by KMT2A-rearranged leukaemia than by everything else? This project already has two executed analyses built on it. The access situation has changed materially and phase-2 plans must account for it: the DepMap portal download API now returns a Cloudflare Turnstile challenge, and DepMap staff have confirmed on the public forum that programmatic download is no longer possible. A no-captcha manifest endpoint exists but the URL column is empty for the recent releases, so the newest data cannot be fetched by script. Two viable routes remain: the Open Targets GraphQL API, which serves per-cell-line gene-effect values and is what this project already uses, and the figshare mirror, which stops at DepMap 24Q4 Public.

Go deeper
What question it answers

For each candidate: mean gene effect across roughly 1,250 lines, gene effect in leukaemia lines, gene effect in KMT2A-rearranged lines, the resulting lineage-controlled selectivity, and co-dependency correlation against MEN1. Every G3 window threshold in the go/no-go register is quoted against these numbers.

How to reach it, in order of preference

1. Open Targets Platform GraphQL (https://api.platform.opentargets.org/api/v4/graphql), field depMapEssentiality { tissueName screens { depmapId diseaseFromSource geneEffect } }. Open, no key, works with plain urllib. This is the route both executed analyses in analyses/ use, and it is the recommended default. Its one defect is that the response does not state the DepMap release version, so every number derived from it carries an unverified provenance 1.

2. figshare mirror, which is complete only up to DepMap 24Q4 Public (2024-12-10) and still serves stable no-login links 2. Use this when a citable, versioned release matters more than currency.

3. The portal, only with a human present. https://depmap.org/portal/api/download/files returns a Cloudflare Turnstile verification page. A no-captcha manifest endpoint returns a 1,436-row CSV listing releases up to DepMap Public 26Q1 (2026-04-01), but the URL column is empty for 25Q2, 25Q3 and 26Q1 - so the manifest tells you what exists without letting you fetch it 2.

The access barrier, stated plainly

This is a new and unannounced-to-us barrier and it has a direct phase-2 consequence: any pipeline that assumes it can refresh DepMap nightly is already broken. The options are to pin to 24Q4, to accept Open Targets values with an unverified release, or to budget a human-in-the-loop captcha step. DepMap has proposed API keys after the 26Q3 release; until those exist, the honest plan is to pin a version and record it.

There is no data-use agreement here - DepMap is open. The barrier is purely mechanical, which makes it easy to under-plan for.

What would make the answer untrustworthy

Two things, both already demonstrated inside this project. First, lineage confounding: a pooled co-dependency correlation can vanish or invert on within-tissue centring, as this project's own MYB claim did (0.105 pooled to -0.052 within tissue). Every DepMap-derived claim in phase 2 must be reported both ways. Second, small-n subgroups: there are only 54 leukaemia lines and single digits of KMT2A::AFF1 lines, so any 'KMT2A-rearranged mean' is an average of a handful of cell lines and should be reported with its n.

Facts and figures
resourceDepMap CRISPR (Chronos) gene effect and Omics
question answeredIs this target more required by KMT2A-rearranged leukaemia than by other cancers, and what else does a menin-dependent cell need?
access barrierNo data-use agreement. A mechanical captcha barrier on bulk download, new as of this review. Pin a version or budget a human-in-the-loop step.
statusplanned
scriptanalyses/depmap-codependency/run.py (existing, reusable)
n leukemia lines54
mandatory reporting ruleevery DepMap-derived correlation must be reported pooled AND within-tissue, because this project has already had a claim overturned by the difference
computelaptop; no GPU
primary access route
{
  "endpoint": "https://api.platform.opentargets.org/api/v4/graphql",
  "auth": "none",
  "field": "depMapEssentiality { tissueName screens { depmapId diseaseFromSource geneEffect } }",
  "proven_in_project": true
}
fallback route
{
  "source": "figshare mirror",
  "latest_complete_release": "DepMap 24Q4 Public (2024-12-10)",
  "auth": "none"
}
blocked route
{
  "endpoint": "https://depmap.org/portal/api/download/files",
  "status": "Cloudflare Turnstile challenge",
  "manifest_endpoint": "https://depmap.org/portal/api/no-captcha/download/files",
  "manifest_rows": 1436,
  "current_release_named": "DepMap Public 26Q1",
  "current_release_date": "2026-04-01",
  "urls_present_for": [
    "24Q2",
    "24Q4"
  ],
  "urls_absent_for": [
    "25Q2",
    "25Q3",
    "26Q1"
  ]
}
existing analyses in project
[
  "ana-depmap-codependency-men1-dot1l",
  "ana-depmap-lineage-conditioned",
  "ana-partner-dependency-vs-essentiality"
]
infant line check 2026 09 19
{
  "candidate": "KOPN-8",
  "cellosaurus": "CVCL_1866",
  "depmap_xref": "ACH-001106",
  "in_depmap_screens_via_open_targets": false,
  "screens_examined": 1258,
  "probe_gene": "MEN1 (ENSG00000133895)",
  "cellosaurus_age_at_sampling": null,
  "disease_label": "B acute lymphoblastic leukemia with t(v;11q23.3) MLL rearranged / Childhood B-ALL",
  "verdict": "The eleven-line no-infant limitation cannot be fixed by adding KOPN-8. The line is catalogued with a DepMap identifier but has no gene-effect data."
}
How this could bite us back
  • DepMap is a bulk proliferation screen in immortalised adult-derived lines over two to three weeks. It is structurally blind to leukaemia-initiating-cell requirements and to fetal-origin cell states, which is exactly the biology of infant KMT2A::AFF1 disease. Every negative DepMap verdict in this beat - CBX8 above all - inherits that blindness.
  • Pinning to 24Q4 to keep reproducibility means phase-2 decisions will be made on data that is more than a year old by the time they are made.
Still open
  1. Does a DepMap 26Q3 release exist, and will the promised API-key scheme restore programmatic access?
  2. How far behind the current DepMap release is the Open Targets copy, and can that lag be measured by comparing a gene with a known recent change?
  3. How many DepMap lines are genuinely KMT2A::AFF1 rather than KMT2A-rearranged generally, and is that number large enough to support any subgroup claim?
  4. Does DepMap model ACH-001106 (KOPN-8, from a 3-month-old infant with KMT2A-r B-ALL) carry Chronos gene-effect data, and if so do the project's selective hits hold in it?
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Per-cell-line DepMap CRISPR gene-effect values are served by the Open Targets Platform GraphQL API without authentication, and this project has already executed two analyses on that route; the API does not state the DepMap release version.

    What this evidence cannot show

    Release version unverified; Open Targets may lag the current DepMap release by an unknown interval, and cell-line disease labels come from a free-text field matched by string search in this project's scripts. TIGHTENED 2026-09-19: the eleven KMT2A-rearranged lines behind every dependency conclusion here have Cellosaurus ages 1Y (THP-1) to 65Y (OCI-AML-2) with HB11;19 unspecified - none infant-derived - while KOPN-8 (CVCL_1866, 3-month-old infant, KMT2A-r B-ALL) carries DepMap id ACH-001106 and is not in the panel. Chronos also retains residual copy-number and screen-quality bias, and the API states no release version.

    Evidence that argues against this (3)
    • The panel is not the population. Every KMT2A dependency conclusion in this project rests on eleven Cellosaurus-annotated KMT2A-rearranged lines, and their ages at sampling, read back from Cellosaurus on 2026-09-19, are THP-1 1Y, SEM 5Y, MV4-11 10Y, MOLM-13 20Y, MOLM-14 20Y, NOMO-1 31Y, RS4;11 32Y, SHI-1 37Y, OCI-AML-2 65Y, with HB11;19 unspecified. The median is adult and none was drawn from an infant under one year - the age group in which KMT2A::AFF1 ALL is most common and most lethal. Cellosaurus (SIB) 2026
    • The gap is not that no infant line exists - it is that this route does not return the one that does. KOPN-8 (Cellosaurus CVCL_1866) was established from a 3-month-old infant with KMT2A-rearranged B-ALL and carries a DepMap cross-reference, ACH-001106. It is absent from the eleven-line panel used here. Whether ACH-001106 carries Chronos gene-effect data at all was NOT confirmed in this check and is the first thing to establish; if it does, the panel should be eleven plus one and the infant line reported separately rather than averaged in. Cellosaurus (SIB) 2026
    • Chronos gene-effect values carry documented systematic artefacts. Meyers et al. established that a gene-independent antiproliferative effect of Cas9 cutting confounds dependency measurement and produces false positives in copy-number-amplified regions, which is why CERES and then Chronos exist at all. Any hit in this project that sits in an amplified region in the rearranged lines needs its copy number checked before it is believed. Meyers RM et al. 2017
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The DepMap portal download API now returns a Cloudflare Turnstile challenge and DepMap staff have confirmed on the public forum that programmatic download is no longer possible; a no-captcha manifest lists releases up to DepMap Public 26Q1 (2026-04-01) but supplies no URLs for 25Q2, 25Q3 or 26Q1, and the figshare mirror is complete only to 24Q4 Public.

    What this evidence cannot show

    Fetched this cycle by a verification worker; access mechanics can change again without notice, and the promised API-key scheme may remove this barrier after the 26Q3 release. Whether a 26Q3 release exists yet is UNVERIFIED.

    Evidence that argues against this (1)
    • Chronos reduces but does not remove the artefacts. Dempster et al. report that Chronos exhibits the lowest copy-number and screen-quality bias among the methods they evaluated - lowest, not zero - and the Open Targets route used here does not state which DepMap release the values come from, so a result cannot be reproduced against a named version or compared across releases. Dempster JM et al. 2021
  3. E_kopn8_absent_from_depmapstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The obvious remedy for 'no infant-derived KMT2A-rearranged line' does not work. KOPN-8 (Cellosaurus CVCL_1866) is a KMT2A-rearranged childhood B-ALL line that carries a DepMap cross-reference, ACH-001106 - but it has NO gene-effect data in DepMap as exposed by the Open Targets Platform. Checked 2026-09-19 against the full screen list returned for MEN1 (ENSG00000133895): 1,258 screens, no ACH-001106 and no cell line whose name matches KOPN. A Cellosaurus DepMap cross-reference records that the line is catalogued, not that it was screened.

    What this evidence cannot show

    Establishes absence from DepMap AS EXPOSED BY OPEN TARGETS, which is the interface every analysis in this project uses; it does not prove absence from a DepMap release file downloaded directly, and a future release could add the line. Also note a correction: KOPN-8 was flagged in-session as derived from a 3-month-old infant, but the Cellosaurus record carries NO age-at-sampling value and labels the disease 'Childhood B acute lymphoblastic leukemia'. The infant characterisation is unverified and is not asserted here.

    Evidence that argues against this (1)
    • This closes one route and not the question. Other KMT2A-rearranged infant lines exist in the literature and were not enumerated here, and absence from Open Targets is not absence from every DepMap artefact. The negative should be read as 'the cheap fix fails', which makes the eleven-line limitation structural rather than an oversight in panel selection - and that is a stronger reason to distrust dependency conclusions for infant disease, not a weaker one. Cellosaurus, SIB 2026
Connected to