Cybergenic
All findings

KRAS ⊥ EGFR

KRAS and EGFR mutual exclusivity in lung adenocarcinoma (TCGA)

Inconclusive: confound not yet controlledMutual exclusivity · Evidence: pattern across patients, not yet tested within subtypes

DataTCGA PanCancer Atlas (2018) · Lung adenocarcinoma · 561 tumours · both altered in 0 (16 expected if independent)

Engine's conclusion

Inconclusive. Significant, but differences between tumour types or subtypes (or, for two genes that sit next to each other on the genome, a single shared amplification) have not been ruled out yet.

A well-studied pair (35,418 papers): on the map it is known biology, a positive control rather than an overlooked discovery.

Statistical support50%Significance (FDR q)0.00055Signal (x)7%Efficacy (y)0%

Signal (x) is the corrected significance on the map's scale, discounted from 38% to 7% because the two genes act in one pathway, so their exclusivity is expected redundancy (known biology), not a new relationship. Efficacy (y) is credibility times actionability.

Structure & site3D structure: AlphaFold

These genes are rarely altered in the same tumour: mutual exclusivity, often a sign they hit one shared pathway so a tumour needs only one broken.

Primary site

Lung

This finding's tumour type arises in the highlighted organ.

Drag to rotate. Dark marks the hotspot residues on the predicted fold. For a fuller read of this same model, open the interactive structure workbench for this protein on Nanoverse (a companion site we also run), which maps its binding pockets, per-region confidence, and the likely effect of a mutation.

KRASOncogeneApproved therapy

KRAS proto-oncogene GTPase

Oncogene: when altered it gets stuck ON and drives growth.

Central RAS signal transducer. G12C is now druggable (sotorasib, adagrasib); G12D inhibitors are emerging.

Pathway
RTK to RAS to MAPK
Hotspots in 3D
G12 · G13 · Q61 · A146
Drives
Pancreatic, Colorectal, NSCLC
KRAS mutations in non-small cell lung cancer →KRAS co-mutations by tumour type →Other findings with KRAS →

Drag to rotate. Dark marks the hotspot residues on the predicted fold. For a fuller read of this same model, open the interactive structure workbench for this protein on Nanoverse (a companion site we also run), which maps its binding pockets, per-region confidence, and the likely effect of a mutation.

EGFROncogeneApproved therapy

Epidermal growth factor receptor

Oncogene: when altered it gets stuck ON and drives growth.

Receptor tyrosine kinase; osimertinib for EGFR-mutant NSCLC, cetuximab in colorectal.

Pathway
RTK to RAS to MAPK
Hotspots in 3D
L858 · T790 · G719 · L861
Drives
NSCLC, Glioblastoma, Head & neck
EGFR mutations in non-small cell lung cancer →EGFR co-mutations by tumour type →Other findings with EGFR →

Anatomical figure from the EBI Expression Atlas anatomogram, and predicted structures from the AlphaFold Protein Structure Database (both CC-BY-4.0). A model shows the protein's healthy fold; a highlighted residue marks a recurrent cancer position, not a prediction of a specific mutation's effect.

Statistical evidence
Test p (raw)
2.9 × 10-7
Effect (log₂ odds ratio)
-5.7
Papers
35,418
Abstracts naming both
3,780
Trials
123
Contingency table: the source of the q561 tumours

The raw 2×2 the FDR q and log₂ odds ratio above were computed from. It is the immutable evidence the score re-derives from: every number on this page traces back to these four counts.

Gene stateKRAS alteredKRAS wild-type
EGFR altered055
EGFR wild-type162344

both 0 (expected 16 if independent) · EGFR-only 55 · KRAS-only 162 · neither 344

Sources
cBioPortal / TCGA PanCancer Atlas (2018)Genome NexusOmniPath / Reactome / STRINGUCSC XenaOpen TargetsPubTator3NIH iCiteEurope PMCClinicalTrials.gov
Anytime-valid evidence (e-value)e = 21

Strong evidence. Exceeds the α=0.05 reject threshold (e≥20). Unlike a single-look p, this stays valid under the engine's continuous re-testing.

Why it sits here on the map

Molecular-subtype certification is unresolved. Control scope was not recorded. Molecular-subtype control is unverified. The association remains visible, but does not qualify as a subtype-controlled discovery.

Opacity: novelty (display only)0%

The two axes are variables: signal (x) and efficacy (y = credibility × actionability). A point high on both lands in the frontier (top-right). A strong signal with low credibility sits low and to the right: statistically strong, not worth pursuing.

KRASOncogeneEGFROncogeneOsimertinibCetuximabSotorasibAdagrasib
Same question in other cohorts
HypothesisModel-generated hypothesis (not evidence)
Written by gpt-6-luna

A language model wrote this from the evidence listed below. It is annotation: the engine's verdict above comes only from the measured statistics, never from this text.

EGFR and KRAS alterations most likely mark alternative dominant-driver states across NSCLC patients; their shared signaling context makes pathway-level redundancy or selection against double-driver tumors plausible, but the available evidence cannot distinguish that from molecular-subtype composition as the main cause of the population-level exclusivity.

Argument strength (model self-report)29%
After adversarial review32%
The argument
  1. 1The depletion is strong: no tumors in the TCGA analysis carry both alterations, the odds ratio is markedly below 1, and the association survives multiple-testing correction. Its recurrence across patient sources supports a real population-level pattern, although it does not establish a mechanism.E1E2E22
  2. 2EGFR and KRAS share pathway context, and the supplied knowledge places both in the RTK–RAS–MAPK pathway. This makes alternative driver states or pathway-level redundancy plausible explanations, but pathway overlap alone does not show that the genes interact or that co-alteration is non-viable within a tumor.E3E16E17
  3. 3Subtype composition remains a serious competing explanation: the observed depletion is across patients, while the provided evidence does not adjust it for molecular subtype. This lead converges with prior Cybergenic findings that highlight subtype composition or alternative driver states in related cancer pairs, especially the NSCLC BRAF–KRAS and EGFR–RBM10 findings.E18E19E20E21
  4. 4The provided paper abstracts do not establish an EGFR–KRAS mutual-exclusivity mechanism: the supplied abstracts describe NSCLC profiling or precision oncology broadly, rather than studying this specific relationship. Although the pair is frequently co-mentioned in the literature, these abstracts do not resolve the mechanism.E12E13E14
What would confirm or refute it

Perform one independently accrued, subtype-annotated NSCLC study using single-cell DNA sequencing to test EGFR and KRAS alteration status in the same cells and assess exclusivity within molecular-subtype strata. Persistence of within-stratum, same-cell depletion would support selection against co-altered clones; disappearance of the association after subtype stratification would favor population composition.

Strongest competing explanation (model)

The strongest deflating explanation is molecular-subtype composition, which has not been resolved here; shared sequencing panels or calling pipelines could also produce systematic artifacts despite cross-cohort consistency. The provided abstracts do not study the EGFR–KRAS relationship, so they do not support a within-tumor cooperation or redundancy mechanism; that specific mechanism remains under-resolved despite broad pair co-mention in the literature.

Adversarial review (second model pass)

Weakest link: The replicated population-level depletion does not establish direct pathway redundancy or selection against double-driver tumors, because the evidence does not adjust for NSCLC molecular-subtype composition or distinguish alteration classes.

Cross-cohort agreement (intra-snapshot)3 sources agree

This mutual exclusivity recurs (same direction, survives FDR) in 3 independent sources of this tumour type. This controls cohort-specific noise only. Several of these cohorts share a sequencing panel, a pipeline or even patients (the MSK cohorts overlap), so a systematic artifact can recur across them. It is not external or prospective validation, and it grants no confidence on its own.

Evidence the argument cites

Model-written annotation, not a finding. The statistics it cites are real and are shown above; the argument connecting them is a hypothesis that needs wet-lab or clinical validation before it means anything.

Cite this finding
Permalink
https://cybergenic.im/findings/0d8b1ac8-b57b-4623-a9cf-70a24e6c2a45
First seen
9 Aug 2026
Prediction registry
Locked 9 Aug 2026, row hash 557f09d6dfdeeec70e5d6415d4146662 (awaiting its out-of-sample test). How the registry works
BibTeX
@misc{cybergenic_0d8b1ac8,
  author       = {{Cybergenic}},
  title        = {{KRAS and EGFR mutual exclusivity in lung adenocarcinoma (TCGA)}},
  year         = {2026},
  howpublished = {\url{https://cybergenic.im/findings/0d8b1ac8-b57b-4623-a9cf-70a24e6c2a45}},
  note         = {Research engine finding, first seen 2026-08-09; prediction locked 2026-08-09, registry row hash 557f09d6dfdeeec70e5d6415d4146662},
  month        = aug
}
See it on the live map Mutations in non-small cell lung cancerEGFR and KRAS in every cohort

Frontier alerts