# Manuscript Outline — EAC dual-arm in-silico discovery-to-design campaign

## Working title
"Steering into competitive whitespace: an in-silico discovery-to-design campaign
nominates GUCY2C and DKK1 and delivers de novo binder leads across the esophageal
adenocarcinoma trajectory"

## Paper brief (derived)
- **Pitch (grandest supportable claim):** A public-data, competition-aware computational
  campaign attacks esophageal adenocarcinoma at two under-exploited points on its
  Barrett's->carcinoma trajectory, delivering de novo protein-binder leads (in-silico
  validated) against a GI-restricted surface antigen (GUCY2C, treatment) and a secreted
  Wnt driver (DKK1, interception).
- **Vision (killer app):** A reproducible, honesty-gated template for going from open omics
  to designed biologic leads while deliberately routing around crowded target space.
- **Audience:** computational protein designers; GI/thoracic translational oncologists;
  target-discovery / drug-hunting teams.
- **Most arresting asset:** Figure 1 — the whitespace map + dual-arm trajectory schematic
  (the strategic thesis in one image).
- **Honest-verdict constraint:** every result is in-silico; no wet-lab. Framing stays at
  "computationally validated leads," not "therapeutics." Two make-or-break questions
  (GUCY2C normal-gut window O-2; DKK1 surrogate-endpoint acceptance O-4) stated as open.

## Figure arc (4 main figures)
1. **Fig 1 — HOOK.** EAC genomics (copy-number driven) + competitive whitespace + dual-arm
   trajectory strategy schematic. Role: hook. One line: "the drivers are crowded; here are
   two open lanes and how we attack them."
   - Panel A: EAC driver landscape (REUSE eac_driver_landscape.png)
   - Panel B: competitive whitespace map (REUSE eac_whitespace_map.png)
   - Panel C: NEW dual-arm trajectory schematic (Barrett's->dysplasia->EAC with DKK1
     interception arm + GUCY2C treatment arm placed on the axis)
2. **Fig 2 — MECHANISM/SELECTION.** Honest druggability triage -> two leads. Role: mechanism.
   - Target-nomination ranked view with the druggability filter made explicit
     (promoted: GUCY2C/CDH17/B7-H3/CEACAM5/DKK1; down-weighted: GPX7, REG4/OLFM4/TFF3).
   - Build from target_nomination.csv if columns support it; else REUSE target_nomination.png.
3. **Fig 3 — EVIDENCE.** De novo binder design + in-silico validation. Role: evidence.
   - Pipeline schematic (RFdiffusion->ProteinMPNN->Boltz-2) + fold-back scatter + per-design
     ipTM bars. REUSE design_results.png (already has scatter + bar); add pipeline strip if needed.
4. **Fig 4 — APPLICATION.** Rendered lead complexes at intended epitopes. Role: application.
   - NEW render: gucy2c_bb2_complex.pdb and dkk1_bb1_complex.pdb; binder vs target chains
     colored, epitope patch highlighted, ipTM/pLDDT annotated.

## Section structure (bioRxiv preprint)
- Title / Authors (AI-assisted campaign; author-contribution stub) / Abstract
- Introduction: EAC unmet need + rising incidence; crowded target landscape; novelty-first
  rationale; trajectory-spanning (interception + treatment) thesis.
- Results:
  R1 EAC is copy-number-driven; drivers crowded, whitespace identified (Fig 1AB)
  R2 Dual-arm trajectory strategy (Fig 1C)
  R3 Target nomination + honest druggability triage -> GUCY2C & DKK1 (Fig 2)
  R4 De novo binder design for both arms (Fig 3)
  R5 In-silico validation and lead complexes (Fig 3, Fig 4)
- Methods: data sources; target-structure prep (AlphaFold DB, domain isolation, epitope
  heuristic); RFdiffusion/ProteinMPNN/Boltz-2 params; triage logic; landscape queries.
- Discussion: what the campaign shows; the two make-or-break open questions; limitations
  (single-model confidence, ipTM as proxy, no wet-lab, GUCY2C bispecific formatting undone,
  epitope heuristic not functional mapping); next steps (ESM/orthogonal fold-back, expression).
- Data & code availability; author contributions; competing interests; references.

## Supplements
- S1: all 16 binder designs (sequences, ipTM, pLDDT, pass) from design_leads.csv
- S2: target nomination table
- S3: competitive landscape table
- S4: detailed computational methods + parameters
- S5: epitope patches + cross-reactivity note
- S6: go/no-go gate table (G1-G5) + open questions (O-1..O-5)