Mosaic is now liveMCP tools, 760+ targets across oncology, neuro and cardio, preprint on Zenodo.See plans →
Mosaic
All theses

Target thesis · Epigenetic writer

EZH2

Histone-lysine N-methyltransferase EZH2

Generated 2026-07-12 · KG version 84 · every number traces to the evidence appendix

Coverage note

read this firstCoverage confidence: Partial

Partial coverage — some dimensions thin or reframed; caveats below.

DimensionHoldingTier
Literature319 curated papers (PubMed ~5,980 oncology)High
Patents113 patents, only 10 organizationsHigh
Compounds3 distinct compoundsLow
Genetic validationpresent (DepMap, 1,208 models)High
Structuralpresent (Foldseek/AlphaFold + AlphaMissense)High

Coverage honesties, up front

  • This thesis does NOT assess chemical-series competition, SAR, or compound selectivity for EZH2 — Mosaic holds only 3 compounds (1 phase-1, 2 phase-4). Any claim about “the EZH2 chemical landscape” would be unsupported by this graph and is therefore not made.
  • Filing velocity is unavailable — there are no parseable filing dates on the 113 patents, so the patent layer cannot tell us whether interest is rising or fading.

Assessed with confidence

  • Genetic validation
  • Complex biology
  • The co-essentiality co-dependency (EED/SUZ12)
  • Competitive concentration (org count)

Flagged / withheld

  • Compound coverage (3 compounds — not assessed)
  • Filing velocity (no parseable dates)

Target-level readiness: Thesis-ready, with a flagged compound-coverage gap.

Headline verdict

Strong non-consensus angle· co-essentiality-backed

Strong complex biology, thin chemical visibility — and one genuinely differentiated, validated read. Mosaic's compound coverage is too thin to judge the EZH2 inhibitor race, so this thesis withholds that claim. What it can commit to: the graph's single best non-consensus signal in the entire test set — real DepMap co-essentiality surfacing EED and SUZ12 (the other two PRC2 complex members) as a validated co-dependency. EED and SUZ12 sit outside Mosaic's curated target set, so the graph holds no patent or compound data on them — their competitive status is not assessed here; the co-dependency is a validated lead to verify externally, an alternative-node angle that direct-EZH2-inhibition discourse misses.

The bull case

  1. 1.

    A validated co-dependency the field under-exploits (see the non-consensus angle). This is the strongest evidence and it is genuinely differentiated.

    confidence: highfunctional-coupling whitespace, DepMap co-essentiality basis

  2. 2.

    Deep literature and a maximally pathogenic hotspot. 319 curated papers; AlphaMissense flags 70.2% of 14,174 variants pathogenic with a hotspot at residue 681 (mean score 1.0).

    confidence: highgenetic_validation.variant_pathogenicity

  3. 3.

    A clinically de-risked modality exists. A phase-4 (approved-stage) EZH2 asset is present in the graph — the mechanism is drugged, not theoretical.

    confidence: moderatecompetitive_landscape (2 phase-4 compounds visible)

  4. 4.

    Concentrated, not saturated, competition. The 113 patents are held by just 10 organizations (Constellation, Repare, and others) — an order of magnitude less fragmented than EGFR (426 orgs) or TIGIT (479).

    confidence: moderatecompetitive_landscape

The bear case

  1. 1.

    Compound coverage too thin to assess — flagged, not papered over. 3 distinct compounds. The chemical-series competition that would normally drive a go/no-go is not visible in this graph.

    confidence: highcoverage.dimensions.compounds

  2. 2.

    Weak pan-model genetic dependency. DepMap mean gene effect -0.049, essential in only 6.9% of 1,208 models. EZH2 is a context-specific dependency (a biological niche), not a broad addiction — efficacy is narrow.

    confidence: highgenetic_validation

  3. 3.

    Resistance durability unknowable. Only 3 curated resistance hits (DOT1L, FGFR4, ...), one paper each — no network read.

    confidence: flaggedresistance_bypass_map

  4. 4.

    Momentum unreadable. No parseable filing dates, so the patent layer cannot tell us whether interest is rising or fading.

    confidence: flaggedcompetitive_landscape

The non-consensus angleStrong non-consensus angle

EED and SUZ12 as a validated co-dependency on the PRC2 complex. Uniquely among the test set, the functional-coupling / co-dependency whitespace tool fires on real DepMap co-essentiality (badge: co-essentiality-backed), not a PPI proxy: EED (r = 0.6974) and SUZ12 (r = 0.6563) — the two other core PRC2 complex members. The CRISPR data says the cells that depend on EZH2 also depend on EED/SUZ12. Coverage caveat: EED and SUZ12 are NOT in Mosaic's curated target set, so the graph holds no patent or compound data on them — their competitive status is not assessed here, and this is explicitly not a whitespace claim. The implication, stated as a validated lead rather than confirmed open space: the PRC2-complex node (EED in particular) is a validated alternative to EZH2 SAM-competitive inhibition; verify its real-world competitive and patent status externally.

  • EED (r = 0.6974) and SUZ12 (r = 0.6563) surface as co-essential with EZH2 on real DepMap CRISPR data — co-essentiality-backed, not a PPI proxy. Both lie outside Mosaic's curated target set, so the graph has no patent or compound coverage for them; their competitive status is not assessed here — a validated co-dependency lead, not a confirmed-whitespace claim.

    confidence: high for the co-essentiality signal; partner competitive status not assessed (out of coverage)functional-coupling whitespace (DepMap-backed badge)

    Patent / compound status: not assessed (out of coverage). gene outside Mosaic curated coverage — competitive landscape is a validated lead to verify externally, not a whitespace claim.

  • Secondary, weaker: a PPI-proxy module (BMI1, DNMT1/3A/3B, and the paralog EZH1) — coupled but not co-essentiality-backed. EZH1 paralog-redundancy is a plausible resistance theme but rests here on PPI only, so it is held at low confidence and not advanced as primary. (These partners are also outside curated coverage; no competitive claim is made on them.)

    confidence: lowfunctional-coupling whitespace (PPI-proxy)

    Patent / compound status: not assessed (out of coverage). gene outside Mosaic curated coverage — competitive landscape is a validated lead to verify externally, not a whitespace claim.

Honest empties

  • Resistance-bypass is thin (3 single-paper hits).
  • Modality cannot be assessed (3 compounds).
  • No emerging-activity surge.

What would change this thesis

  • External verification of EED/SUZ12 competitive and patent status — Mosaic does not cover these genes, so the graph cannot confirm whether the PRC2-complex node is genuinely open. This external check is the single most load-bearing condition on the non-consensus angle.
  • If more EZH2 compounds are ingested, the compound dimension becomes assessable and may reveal the inhibitor race is in fact crowded — which would add a bear point currently withheld.
  • If EED/SUZ12 co-essentiality fails to replicate in an EZH2-gain-of-function-selected model subset, the non-consensus angle weakens from “validated” to “hypothesis.”
  • If a context is shown where EZH2 dependency is high, Bear #2 (weak pan-model dependency) softens.