# Plain-Language Summary

**What is this about?**
Eosinophilic esophagitis (EoE) is a chronic allergic disease in which the immune system reacts to food and inflames the esophagus, the tube connecting the throat to the stomach. Every approved treatment works by broadly damping down inflammation and must be taken indefinitely — stop, and the disease returns. None of them fixes the actual root cause, and none teaches the immune system to stop over-reacting for good ("tolerance"). Deep down, EoE is an antigen-specific disease: a food protein fragment is shown to the immune system by a molecule called HLA (MHC-II), which sets off the reaction. The hard part is that different patients react to different foods (milk, wheat, egg, soy — often several at once), so there is no single fragment to target for everyone.

**What did we do?**
Using only public data and computer methods — no laboratory experiments — we built an "antigen-selection engine." It takes a patient's individual HLA genetic type and calculates which food-protein fragments that particular person's immune system is able to display. We then designed a way to place a chosen fragment onto a standard molecular scaffold and mount several copies on a tiny nanoparticle, a format intended to retrain the immune system toward tolerance rather than attack.

**What did we find?**
Scanning three major EoE food allergens (milk, wheat, soy) across five common HLA types, the engine predicted 832 strong binders out of 4,640 combinations tested. Crucially, how many fragments a person can display depends heavily on their HLA type — ranging from 13 to 352 across the same set of foods. That is the key insight: which food matters most is genuinely individual and must be computed per patient, not assumed. As a worked example, we advanced a milk fragment matched to one common HLA type; the AI-predicted 3D structure passed our quality bar with the fragment sitting correctly in its groove, and we specified a 20-nanometre particle carrying five copies to boost its effect roughly 24-fold.

**How sure are we?**
This is early, computer-based work. Every fragment reported is a prediction — no laboratory tetramer or T-cell tests have been done yet, and the predicted structures are model estimates, not experimental ones. We validated the structural step for only one HLA type so far; extending it to all five types and to more allergens (like egg) remains to be done. This kind of tolerance therapy is also still clinically unproven in any disease. We lay out a step-by-step lab validation plan and are explicit that this is a design framework, not medical advice.

**Why does it matter?**
EoE's biggest obstacle — that patients react to different, multiple foods — is exactly what has made a targeted cure seem out of reach. By turning antigen selection into a per-patient calculation, this work reframes that obstacle as a solvable, personalized-medicine problem, and offers a reproducible template that could extend to celiac disease and other food- or self-antigen conditions.