# Executive Summary — EoE Convergent-Target De Novo Binder Preprint

**Paper:** *De novo design of neutralizing protein binders against CCL26 and POSTN, two convergent targets in eosinophilic esophagitis.*
**Status:** bioRxiv-style preprint. **All results are computational (in-silico); no experimental validation has been performed.**

## What the paper does
It converts a public-data target-discovery and protein-design campaign into a submission-ready preprint. Working from public data alone — a nine-cohort EoE transcriptomic meta-analysis plus literature and clinical-pipeline mining — it (1) shows that approved and late-stage EoE therapies all converge on a small set of upstream nodes (IL-4/IL-13, IL-5/IL-5Rα, TSLP), (2) nominates two reproducibly dysregulated targets with no direct therapeutic, and (3) carries out structure-based de novo binder design against a defined functional epitope on each — with an explicit negative-control calibration that is itself a central result.

## The two leads
- **CCL26 / eotaxin-3** (anti-chemokine arm) — the eosinophil-selective chemokine that is the single most discriminating EoE transcript (PMID 17900656), proposed for a ligand-trap that neutralizes recruitment. Lead binder engages the CCR3/glycosaminoglycan face: ipTM 0.942, 4/5 epitope hotspots, 1,220 Ų buried.
- **POSTN / periostin** (anti-remodeling arm) — a matricellular remodeling protein that stays elevated in histologic remission alongside persistent DSG1 loss (PMID 39343172), marking residual disease current therapies leave behind. Lead binder engages the FAS1 integrin-interaction patch: ipTM 0.901, complex pLDDT 0.911, 3/4 hotspots, 842 Ų buried.

## Key numbers (all traceable to artifacts/public sources)
- Discovery: nine-cohort transcriptomic meta-analysis; both targets strongly and reproducibly dysregulated, neither addressed by any approved or late-stage agent.
- Design campaign: 80 RFdiffusion backbones → 1,920 SolubleMPNN sequences → 60 binder–target complexes assessed by Boltz-2 co-folding.
- **Calibration (the cautionary result):** against 36 scrambled/random-sequence decoy complexes folded under the identical protocol, 92–100% of decoys cross the same ipTM > 0.5 / complex-pLDDT > 0.7 thresholds that 59/60 designs cross — so a high pass rate is close to the null. The only readout separating designs from decoys is focal on-epitope engagement, and that signal is modest.
- Interfaces map to the CCR3/GAG face of CCL26 and the FAS1 patch of POSTN.

## Why it matters
It targets two biologically central, pharmacologically unexploited axes in EoE — eosinophil recruitment and tissue remodeling — that lie downstream of every current drug, and delivers concrete designed binders a wet-lab program could take forward. Equally, its decoy calibration is a transparent, reusable lesson on what in-silico co-folding confidence can and cannot establish for small compact targets.

## Honest verdict and what remains
The framing matches the evidence: these are in-silico leads, not validated therapeutics, and the paper's own decoy experiment is an explicit warning against over-reading confidence scores. Two biological cautions are carried openly — eotaxin redundancy (CCL11/CCL24 also signal through CCR3, so trapping CCL26 alone may be partially bypassed) and periostin's physiological role in tissue repair (an on-target liability a biomarker does not carry). Experimental validation — recombinant expression, binding measurement, and functional neutralization — is the required next step, not a claimed one.