======================================================================
NF1-Loss Pan-Cancer Dependency Atlas
FINAL SYNTHESIS AND RECOMMENDATION
Analyst: analyst | Division: cancer | Date: 2026-03-24
======================================================================

======================================================================
SUCCESS CRITERIA SCORECARD
======================================================================

CRITERION 1: >=5 NF1-loss-specific dependencies with |d|>0.5 and
FDR<0.1 in >=2 cancer types
  STATUS: NOT MET
  Result: 1 hit (RIT1, d=-0.646, FDR=0.071, pan-cancer only — does
  not appear in >=2 individual cancer types at FDR<0.1).
  Zero per-cancer-type hits survive FDR correction.
  Sub-threshold coherent gene set (SHOC2, SPRED1, SPRED2 at FDR 0.2-0.3)
  brings total to ~4 credible candidates, still below 5.

CRITERION 2: At least 2 hits with druggable targets (clinical-stage
or FDA-approved)
  STATUS: PARTIALLY MET (with caveats)
  Credibly supported druggable targets:
    1. SHP2/PTPN11: Phase 2 validated (d=-0.389, FDR=0.017 in MAPK
       baseline). TNO155, RMC-4630 in Phase 1/2 trials.
    2. RAF1/C-RAF: Phase 2 borderline (d=-0.368, FDR=0.052).
       LY3009120, Belvarafenib (pan-RAF, Phase 1/2).
  Both are sub-threshold in genome-wide screen (FDR>0.2) but validated
  in the targeted Phase 2 analysis with appropriate multiple testing
  correction. Credible but marginal.
  CDK6, ATR, CHEK1, WEE1 have drugs but DepMap CRISPR support is
  too weak or TP53-confounded to count.

CRITERION 3: Clear differentiation from KRAS-mutant dependency
profiles (>=3 NF1-specific hits not shared with KRAS)
  STATUS: PARTIALLY MET (quality concerns)
  Phase 5 identified 259 NF1-specific genes vs 1 shared dependency.
  However, the NF1-specific list is artifact-contaminated (top hits:
  KRTAP21-1, SPRR1A = lineage bias; USP9Y, EIF1AY = sex bias).
  After filtering artifacts, credible NF1-specific hits include:
    RIT1, SHOC2, SPRED1, SPRED2 (at least 3 after artifact removal).
  At the PATHWAY level, NF1 vs KRAS differentiation is clearer:
    NF1: GRB2/SHP2/SHOC2 dependent, MEK-independent
    KRAS: RAF1/MEK dependent
  This qualitative distinction is the strongest finding of the atlas.

CRITERION 4: MPNST-specific findings with therapeutic potential
  STATUS: NOT MET
  MPNST top DepMap hits (SH2D6, FKBP6, TGM6) are undruggable and
  biologically implausible (poorly characterized genes, FDR>0.9,
  n=8 vs n=27).
  Recommended MPNST candidates (MEK, mTOR, CDK4/6, SHP2, BRD4, EZH2)
  are literature-derived, not DepMap discoveries. Only SHP2 has direct
  CRISPR support from this atlas. MEK is actually counter-indicated
  by our data (MAP2K1 d=+0.265).

======================================================================
OVERALL SCORECARD
======================================================================

  Criterion 1 (>=5 SL hits):           NOT MET (1/5)
  Criterion 2 (>=2 druggable targets): PARTIALLY MET (2, with caveats)
  Criterion 3 (NF1 vs KRAS):           PARTIALLY MET (quality issues)
  Criterion 4 (MPNST findings):        NOT MET

  OVERALL: 0/4 fully met, 2/4 partially met, 2/4 not met.

======================================================================
KEY CAVEATS
======================================================================

1. TP53 CONFOUNDING (Critical)
   71.5% of NF1-lost lines are TP53-mutant. No TP53-stratified
   analysis was performed. All DDR, cell cycle, and some epigenetic
   signals cannot be attributed to NF1 vs TP53. Developer task #1335
   created to address this.

2. SAMPLE SIZE LIMITATIONS
   Per-cancer-type analyses (n=5-18) are fundamentally underpowered
   for genome-wide discovery. The study design was appropriate for
   pan-cancer pooling only.

3. HIGH FDR BURDEN
   160,947 tests with BH-FDR means only extreme signals survive
   correction. The "negative" genome-wide result may partly reflect
   insufficient power rather than absence of biology.

4. PRISM COVERAGE GAP
   Key drug classes (MEKi, mTORi, CDK4/6i, SHP2i) absent from
   PRISM 24Q2 dataset. Pharmacological validation was not possible.

5. MPNST CELL LINE SCARCITY
   n=4 NF1-LOF nerve sheath tumors with CRISPR data is insufficient
   for any statistical inference about MPNST-specific dependencies.

6. DISCONNECT WITH PUBLISHED LITERATURE
   Published in-vivo NF1-SL data (mTORi, CDK4/6i synergy with MEKi)
   is NOT supported by DepMap CRISPR data. This limits the atlas's
   ability to extend beyond known biology.

======================================================================
WHAT THE ATLAS DID FIND (Credible Positive Results)
======================================================================

Despite not meeting formal success criteria, the atlas produced one
coherent and novel biological finding:

** NF1-LOSS CREATES NON-CANONICAL MAPK PATHWAY DEPENDENCY **

NF1-lost cells depend on upstream RAS regulatory nodes (GRB2, SHP2,
SHOC2) and parallel RAS-family GTPases (RIT1, SPRED1/2 as negative
feedback regulators) — NOT on downstream effectors (MEK1, ERK1/2).

This is mechanistically distinct from KRAS-mutant dependencies
(which are RAF1/MEK-centered) and explains clinical observations:
  - MEK inhibitor monotherapy fails in MPNST (ORR<10%)
  - SHP2 inhibitors are being tested in NF1-loss contexts
  - MEK+SHP2 combos are rationally designed for NF1 (SHP2 upstream
    of the NF1-regulated RAS node)

This pathway-level finding, while not meeting individual gene-level
success criteria, is the atlas's most valuable contribution.

======================================================================
RECOMMENDATION
======================================================================

RECOMMEND: FLAG FOR RESEARCH DIRECTOR REVIEW with specific guidance.

Do NOT advance to full validation — the dataset does not support
a standard validation pipeline (too few significant hits). Instead:

1. ADVANCE TO LIMITED VALIDATION (specific scope):
   a) SHP2/PTPN11 + RAF1: These have Phase 2 CRISPR support and
      druggable targets. Validate with targeted CRISPR experiments
      in NF1-mutant cell lines (confirmatory, not discovery).
   b) RIT1: Single genome-wide hit. Validate dependency in NF1 vs
      WT isogenic pairs if available.
   c) SPRED1/2 synthetic lethality: Biologically novel finding.
      Validate in NF1-isogenic systems. Could be publication-worthy
      even without drug target.

2. RETURN TO DEVELOPER for critical improvement:
   a) TP53-stratified analysis (task #1335 already created).
   b) Lineage-adjusted analysis (remove skin/keratin bias from
      pan-cancer pooling, or add lineage as covariate).
   c) Pathway-level testing (GSEA/CAMERA for RAS regulatory gene set)
      to improve power vs individual gene testing.

3. DO NOT REPORT as publication-ready:
   a) Per-cancer-type hit lists contain >50% artifacts.
   b) MPNST recommendations are literature-derived, not atlas-derived.
   c) PRISM analysis is uninformative due to coverage gaps.
   d) NF1-vs-KRAS comparison needs artifact filtering before reporting.

4. SCIENTIFIC NARRATIVE for research_writer:
   If proceeding to documentation, frame as:
   "NF1-loss creates a distinct non-canonical MAPK dependency profile
   centered on upstream regulatory nodes (SHP2, GRB2, SHOC2, SPRED1/2)
   rather than canonical MEK/ERK effectors, providing mechanistic
   rationale for SHP2 inhibitor combinations in NF1-mutant cancers."
   This is the defensible narrative — not "we found X novel drug targets."

======================================================================
FILES PRODUCED
======================================================================

output/cancer/nf1-loss-pancancer-dependency-atlas/analysis/
  phase1_2_review.txt            — Classifier quality & MAPK baseline
  phase3_assessment.txt          — Genome-wide screen critical review
  phase4_5_evaluation.txt        — Drug targets, PRISM, cross-atlas
  synthesis_and_recommendation.txt — This document
