The papers we build on, and what to do with them.

What it found

Children who ate peanut from infancy to age five were followed to age twelve. At 144 months, peanut allergy was present in 4.4 percent of the early-eating group and 15.4 percent of the avoidance group. Protection held even though many children went years without eating peanut.

Why it matters here

This is the clearest existing proof that a cheap, early, non-pharmaceutical intervention can produce durable immune tolerance. It is the benchmark for what we mean by long-lasting and affordable.

What it found

A four-arm trial in Norway and Sweden tested introducing peanut, milk, wheat and egg from three months of age, applying skin emollients from two weeks, both, or neither. Early food introduction reduced food allergy at three years. The emollient arm did not.

Why it matters here

A negative result published as loudly as a positive one. It tells us that feeding the gut and greasing the skin are not interchangeable, and it is a model for the honesty we want in our own reporting.

What it found

An anti-IgE antibody given for 16 to 20 weeks let 67 percent of treated children tolerate 600 mg of peanut protein without dose-limiting symptoms, against 33 percent on placebo, with protection extending to other foods at the same time.

Why it matters here

It treats several allergies at once by hitting a shared mechanism rather than one food. That is exactly the leverage we look for. It is also expensive and must be re-dosed, which is exactly the problem we exist to solve.

What it found

A wearable peanut patch worn daily for twelve months raised the reaction threshold in 67 percent of one-to-three-year-olds against 33.5 percent on placebo. Adverse events occurred in essentially every child in both groups, most of them local skin reactions.

Why it matters here

Delivery route is a design choice with enormous consequences for cost, cold chain, and whether a family in a rural clinic can use the thing at all. A patch is a template for access-first design.

What it found

Peanut oral immunotherapy in toddlers was tested for desensitization at 134 weeks and, critically, for remission after a further 26 weeks off treatment. Separating those two endpoints is the point of the trial.

Why it matters here

IMPACT is where the field learned to distinguish protection while on a drug from tolerance that survives stopping it. Program P-02 is built on that distinction.

What it found

An argument that detergents, surfactants, microplastics, pollutants and processed food damage the skin, gut and airway linings, and that this damage drives the global rise in allergic and autoimmune disease.

Why it matters here

If barrier damage is upstream of the disease, prevention means fixing the barrier rather than suppressing the immune response downstream. This single idea shapes our whole target-selection logic.

What it found

Children who went on to develop food allergy had fewer Bifidobacterium and Clostridia species and a less mature infant microbiome. The protective effect appears to run through anti-inflammatory metabolites and regulatory T cells.

Why it matters here

It names a window, roughly conception to age two, in which prevention is plausible and cheap. A foundation with no bench can still do serious work on when to intervene.

What it found

Topically applying a harmless skin bacterium to fifteen children with eczema was associated with reduced severity, better barrier function, less Staphylococcus aureus, and lower steroid use. Improvement persisted up to eight months after treatment stopped.

Why it matters here

A living organism, grown cheaply, with an effect that outlasts dosing. That is close to the shape of the intervention we are trying to find, and the authors are careful to call it a signal, not proof.

What it found

Innate immune cells, long assumed to have no memory, can be durably reprogrammed through metabolic and epigenetic changes so they respond more strongly to later challenges.

Why it matters here

Long-lasting immune change without a long-lasting drug. If a single exposure can rewire innate cells for months, durability becomes a biological property rather than a formulation problem.

What it found

A single diffusion-based model predicts the joint structure of proteins together with nucleic acids, small molecules and ions, beating specialized docking tools at protein-ligand prediction and earlier methods at antibody-antigen prediction.

Why it matters here

Structural biology used to be the expensive gate between an idea and a testable hypothesis. It is now largely a compute problem, which is why a foundation this small can run a discovery program at all.

What it found

An open-source model reaching AlphaFold 3-level accuracy on biomolecular complexes, released with training code, inference code, weights and benchmarks under the MIT license.

Why it matters here

Open weights are the difference between reading about a method and running it. Everything we build has to be reproducible by a lab that cannot sign a commercial license.

What it found

A diffusion model generates new protein backbones to order, including binders designed against a chosen target. Hundreds of designs were made and experimentally characterized, and a cryo-EM structure of one binder matched its design model almost exactly.

Why it matters here

Designing a protein that blocks a specific allergic pathway is no longer science fiction. The expensive part is now synthesis and testing, which is precisely where a partner lab earns its place.

What it found

A graph neural network trained on a medical knowledge graph ranks existing drugs as candidate treatments for 17,080 diseases, including diseases with no approved therapy, and explains each prediction as a path through the graph.

Why it matters here

Repurposing an approved drug skips most of the $2.6 billion and most of the decade. For a foundation chasing affordability, it is the single highest-leverage computational method in this list.

What it found

Generative models screened over ten million chemical fragments in silico and proposed new antibacterial compounds. Of 24 compounds synthesized, seven were selectively antibacterial and two cleared multidrug-resistant infections in mice.

Why it matters here

A full loop from generated molecule to animal efficacy, reported with its real hit rate. Seven out of twenty-four is the honest number that tells you what AI-first discovery actually costs.

What it found

A free, open-source knowledge base linking genes to diseases, with a target prioritization view scoring clinical precedence, tractability, doability and safety, and an assessment of whether a variant's effect raises or lowers risk.

Why it matters here

This is the backbone of Program P-01. Target ranking used to require a pharmaceutical company's data licenses. It now requires a browser.

What it found

A public archive holding over 200,000 studies and 6.5 million samples, now with consistently computed gene expression count matrices for thousands of RNA-seq studies and browser-based differential expression tools.

Why it matters here

Decades of immunology experiments, already paid for, already public, mostly unreanalyzed. Our entire first program rests on the premise that this data has more in it than has been taken out.

What it found

A collection of curated datasets, standardized tasks and leaderboards spanning the stages of drug discovery, built so that competing machine-learning methods can be compared on identical footing.

Why it matters here

It answers the question that stops most newcomers: which method should I use. Shared benchmarks are how a field stops arguing and starts improving.

What it found

A practical guide to running machine-learning studies in pediatric allergy, covering core concepts, how to structure an analysis pipeline from variable selection to interpretation, and the common pitfalls. It notes that most work so far has used simple single-source data and reported its methods poorly.

Why it matters here

The best entry point in this list. It is the paper we would hand a new volunteer on their first day, and its critique of weak methodology is the standard we hold ourselves to.