KRAS

The bull case

1. 1.

   Selective genetic addiction — the real thing. DepMap mean gene effect -0.7249, essential in 52.4% of 1,208 models (dependency fraction 0.524). This is a true selective dependency, materially deeper than EGFR's 17.6%.

   confidence: high·genetic_validation

2. 2.

   Strongly pathogenic, hotspot-driven mutational landscape. AlphaMissense flags 74.5% of 3,591 scored variants pathogenic, with a hotspot at residue 57 (0.9998).

   confidence: high·genetic_validation.variant_pathogenicity

3. 3.

   Deep, recent commercial conviction. 416 patents; filing velocity surged 9 (2021) → 24 (2022) → 112 (2023) and held at 106 in 2025. Named leaders (Amgen 21, Mirati 15, Astellas 14) signal a real race, not speculative noise.

   confidence: high·competitive_landscape

4. 4.

   Literature depth. 366 curated papers — the densest target in the KG, supporting a well-evidenced view.

   confidence: high·coverage.dimensions.literature

The bear case

1. 1.

   Still overwhelmingly pre-clinical. Of 94 compounds, 86 are preclinical (phase 0); the clinic now shows 4 phase-1, 1 phase-2, a phase-3 and 2 phase-4 — early clinical entrants have appeared, but the franchise is unproven at scale.

   confidence: high·competitive_landscape.compound_phase_distribution

2. 2.

   The gold-rush has crowded it. 416 patents / 349 organizations with a 2023 spike of 112 filings — direct G12C/pan-KRAS chemical space is contested.

   confidence: high·competitive_landscape

3. 3.

   Resistance is already documented — dominated by on-target escape. The strongest signal is KRASG12C itself, 41 papers across 33 independent sources; downstream bypass nodes (SOS1, RAF1, MAP2K1) are already drugged elsewhere (96–156 compounds) but, per source-doc dedup, largely co-derive from shared reviews rather than independent evidence. Durable monotherapy is unlikely.

   confidence: high·resistance_bypass_map + dedup

4. 4.

   Modality narrowness. The only typed chemical matter is 2 covalent compounds; the KG cannot evidence a non-covalent or pan-RAS small-molecule franchise.

   confidence: moderate·kg_native.modality_gaps (flagged; classifier covers ~2%)

The non-consensus angleModerate non-consensus angle

On-target resistance is the real, independently-corroborated signal. Of the curated resistance hits, only KRASG12C clears the bar: 41 resistance-context papers across 33 distinct source documents (acquired + general). The implication is a combination-first posture — durable monotherapy is unlikely. This is well-sourced but not co-essentiality-grade, so it is framed as directional.

• On-target resistance is the real, independently-corroborated signal: KRASG12C, 41 resistance-context papers across 33 distinct source documents. The implication is a combination-first posture — durable monotherapy is unlikely.

  confidence: headline-eligible — 33 independent sources·kg_native.resistance_bypass_map (curated)

• The downstream “bypass network” is thinner than node-counts suggest (dedup caveat). The raw map lists 18 curated nodes, but they trace to 46 source docs with heavy sharing — one review alone supplies 5 nodes (FGFR3, MAP2K1, RAF1, RET, KRASG12C). FGFR3, RAF1, MAP2K1, WEE1, BRD4 and others are co-derived from shared reviews with no node-unique evidence. Only SOS1 (2 papers, 1 node-unique; already drugged) has standalone downstream support.

  confidence: source-doc dedup·synthesis_input...network_dedup

• RAS-effector module — suggestive, not validated. After excluding 5 calmodulin PPI artifacts, the functional-coupling whitespace tool leads with RALGDS, RGL1, NF1, RASSF1, SOS2 — a coherent RAL-effector / RAS-GAP module distinct from the crowded RAF/MEK axis. None is in Mosaic's curated target set, so the graph holds no patent/compound data on them; their competitive status is not assessed (not a whitespace claim). And this is PPI-proxy only — no DepMap co-essentiality fired — so it is a weaker observation worth checking externally, not a co-essentiality-grade finding.

  confidence: PPI-proxy — low-moderate·kg_native.synthetic_lethal_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.

What would change this thesis

• If a phase-2/3 readout shows a direct KRAS inhibitor holds durable monotherapy response, Bear #3 (resistance-bound) weakens and the combination-first thesis softens.
• External verification of RALGDS / RGL1 / NF1 competitive status — these genes are outside Mosaic's curated coverage, so the graph cannot confirm the effector module is open; verify externally before treating it as whitespace.
• If DepMap co-essentiality data resolves a true synthetic-lethal partner for KRAS (none fired here), the non-consensus angle upgrades from PPI-proxy to validated.
• Bear #1 has already begun to soften: the phase distribution has matured since the prior snapshot — a phase-3 and a phase-2 entrant now sit above the preclinical wave (86 of 94 still phase 0). A positive phase-3 readout would weaken it further.