Oncofusion

start / graph / exp-guide-level-decomposition-of-the-cbx1-depmap-screen

Experimentmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q35-encode-reanalysis

Decompose the CBX1 DepMap screen to single guides before anyone quotes -0.349 as biology: a no-wet-work, one-afternoon check that would either kill the HP1 paralogue asymmetry or make it the only HP1 question in this graph with a phenotype

also called the CBX1 guide check · excluding paralogue cross-reactivity

Before believing that cancer cells specifically need one of three near-identical proteins, check whether the gene-editing tool used to test it was accidentally cutting the other two as well. The data to check are already public.

The same thing, for a scientist

**Why this is the first thing to do and not the tenth.** This graph is about to carry a claim - CBX1 is four times the dependency of CBX3 or CBX5 - that is quoted in `data/symbol-collisions.json` as evidence the paralogues "are not interchangeable". **The figure comes from an aggregated gene-effect score that cannot, in principle, distinguish a gene-specific requirement from a guide that cuts more than one member of a three-member family within eight residues of each other.** The distinguishing data are public and the check costs no wet work. **Design.** Download the DepMap public CRISPR (Avana/Chronos) per-guide log-fold-change table and the guide-to-gene map for the same release the Open Targets Platform 26.06 payload is derived from. For every guide annotated to CBX1: (1) compute its own median effect across the 1,257 lines; (2) align its 20-mer protospacer plus PAM against the CBX1, CBX3 and CBX5 transcripts and the whole genome, recording perfect and one-mismatch hits; (3) recompute the CBX1 gene-level mean after dropping every guide with a family-internal off-target. Run the identical pipeline on CBX3 and CBX5 as controls, and on a gene with a known clean single-copy target as a positive control for the alignment step. **Readout and pre-stated thresholds** - these are the falsifier thresholds of `hyp-the-hp1-paralogue-that-costs-fitness-is-the-one-this-field-never-measures` and are repeated here so the experiment is scoreable on its own: the asymmetry is **artefactual** if dropping cross-cutting guides lifts the CBX1 mean above -0.15; **real** if at least 80% of CBX1 guides individually reach below -0.25 with no family-internal off-target; **unresolved** in between. **What makes this worth a beat's time rather than a footnote.** It is the cheapest available test of a number this graph has already written into a registry row, and both outcomes are publishable inside the graph: an artefact finding corrects the registry and is a reusable warning about every paralogue-family DepMap comparison this project makes, of which there are several. A real finding makes CBX1 the only HP1 paralogue with a fitness phenotype and forces the DSB cascade to argue for its choice of HP1-gamma.

Facts and figures
designPer-guide decomposition of the DepMap public CRISPR screen for CBX1, with CBX3 and CBX5 as matched controls: per-guide median effect across 1,257 lines; protospacer alignment against the three paralogue transcripts and the genome; gene-level mean recomputed after dropping guides with family-internal off-targets.
readoutCBX1 gene-level mean gene effect before and after dropping cross-cutting guides; the per-guide effect distribution; the count of CBX1 guides with a perfect or one-mismatch hit in CBX3 or CBX5.
controlsCBX3 and CBX5 run through the identical pipeline (if the effect is artefactual in one direction it should be visible in the others); a clean single-copy gene as a positive control for the alignment step; and the aggregated Open Targets figure reproduced from the per-guide table as a sanity check that the right release was downloaded.
cost tierlow
timelineone afternoon; no wet work, no institutional access beyond the DepMap public download
what it cannot doIt cannot say anything about HP1 paralogue function at the KMT2A locus, which this cycle measured separately and found negative for all three in the resting state, nor about the damage state. It bounds one specific artefact and nothing else.
why the endpoint this project uses cannot answer itThe Open Targets Platform GraphQL depMapEssentiality field returns aggregated per-cell-line gene effects and no guide-level data at all. Every DepMap number in this graph comes through that endpoint, so no existing entry in this corpus could detect a guide artefact in any gene.
shown for fusionNO - this is a pan-lineage screen question, not a fusion experiment
preregistered thresholds
{
  "artefact": "CBX1 mean rises above -0.15 after dropping cross-cutting guides",
  "real": ">=80% of CBX1 guides individually below -0.25 median AND zero family-internal off-targets",
  "unresolved": "anything in between, recorded as unresolved rather than as support"
}
Still open
  1. Should this check be run over every paralogue-family DepMap comparison in this corpus rather than only CBX1? The graph makes several - RAS family, HDAC family, CBX family - and the same defect would apply to all of them.
  2. Which DepMap release does Open Targets 26.06 actually wrap? The payload carries no release id (rule 83), so matching the per-guide download to it requires an assumption, and getting that wrong would make the whole comparison invalid.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The Open Targets Platform depMapEssentiality endpoint, which is the route every DepMap figure in this graph comes through, returns aggregated per-cell-line gene effects and exposes no guide-level data, so no entry in this corpus can currently detect a guide cross-reactivity artefact in any gene.

    DepMap CRISPR gene-effect (Chronos) for CBX1, served by the Open Targets Platform GraphQL API · Broad Institute DepMap / Open Targets Platform · Open Targets Platform 26.06 (meta.dataVersion year 26 month 06), fetched 2026-09-22 2026
    What this evidence cannot show

    This is a statement about the fields this beat requested and received from that endpoint on 2026-09-22; a different query shape might expose more, and that was not exhaustively tested (rule 93 applies to this claim as much as to any other).

    Evidence that argues against this (1)
    • The aggregated score is not worthless - Chronos already models copy-number effects and multi-target cutting to some degree, so the artefact this experiment tests for may be partly corrected already. The experiment is still worth running because the correction is for multi-cutting in general, not for the specific case of near-identical paralogues. Broad Institute DepMap / Open Targets Platform 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    CBX1, CBX3 and CBX5 are 185, 183 and 191 residues with shared chromo and chromo shadow domains, which is the paralogue similarity that motivates the check.

    What this evidence cannot show

    Protein-level similarity, not nucleotide-level guide-target identity, which is what the experiment would actually compute.

    Evidence that argues against this (1)
    • If nucleotide divergence between the three is high despite protein similarity, the alignment step will return zero family-internal off-targets immediately and the experiment ends in twenty minutes with the artefact excluded. That is a good outcome, not a failed experiment. UniProt Consortium 2026
Connected to