# Plain-Language Summary

**What is this about?**
Eosinophilic esophagitis (EoE) is a chronic allergic disease of the esophagus — the tube that carries food to the stomach. In EoE, the immune system reacts to food, and a type of white blood cell called an eosinophil floods into the esophagus, causing inflammation, scarring, and difficulty swallowing. Today's newest medicines all work at the same early step of this reaction (the cytokine "alarm signals"), which leaves two other important parts of the disease with no dedicated treatment.

**What did we do?**
Using only publicly available scientific data and computer methods — no laboratory experiments — we combined a meta-analysis of nine EoE patient study groups with a review of the drug pipeline to look for proteins that are strongly and repeatedly abnormal in EoE but that no current or late-stage drug targets. We then used AI protein-design tools to try to build brand-new "binder" proteins — small custom molecules designed to latch onto each target and switch it off.

**What did we find?**
We nominated two overlooked targets, each at a different point in the disease:
- **CCL26 (eotaxin-3)** — the specific signal that actually calls eosinophils into the esophagus. It is the single most distinctive gene separating EoE from ordinary acid reflux, yet nothing on the market blocks it directly.
- **POSTN (periostin)** — a protein that drives the tissue scarring and stays elevated even when patients look like they are in remission, marking damage that current drugs leave behind.

We designed candidate binders against a functional "business end" of each protein. Just as importantly, we ran a careful honesty check: we folded scrambled, meaningless decoy sequences through the exact same pipeline. Almost all of the decoys passed the usual computer confidence scores just as easily as our real designs did — a cautionary result showing those scores alone do not prove a design works. The one measure that did separate real designs from decoys was whether the binder actually grabbed the intended spot on the target, and by that measure two leads stood out.

**How sure are we?**
This is early, computer-based work. Nothing has been made or tested in a lab, and we say so throughout. Our own decoy experiment is a built-in warning that in-silico confidence scores can be misleading for small targets like these. There are also biological cautions: other closely related proteins can partly substitute for CCL26, and periostin has normal jobs in healthy tissue repair, so blocking it carries risk. These are the questions that lab work will need to answer.

**Why does it matter?**
The work points to two genuinely different, unexploited ways to treat EoE — shutting off the recruitment signal for eosinophils, and calming the scarring that persists after inflammation is controlled. Equally valuable, it provides an unusually candid, reproducible recipe — including what AI protein-design scores can and cannot tell you — that other scientists can build on at the bench.