Every Cure

David Fajgenbaum nearly died five times from a rare immune disease before he found his own cure on a pharmacy shelf: sirolimus, a transplant drug never meant for his condition. The repurposing search that saved him was done by hand, on one patient. His nonprofit, Every Cure, has now built the machine version.

A few hundred patients a year in the US, Fajgenbaum estimates, hit the kind of deadly flare the model's drug pulled one man out of.

Its platform scores all roughly 75 million possible pairings of about 4,000 approved drugs against some 18,000 diseases, ranking which existing medicines might treat conditions they were never designed for. The full scoring run, Every Cure says, has gone from about 100 days three years ago to roughly 17 hours. The numbers are the company's own and no outside group has audited the predictions' accuracy, but the commitment behind them is unusual: most firms doing this keep their score tables proprietary, and Every Cure plans to release the entire ranked list publicly.

Knowledge-graph drug repurposing is not new — it flagged an arthritis drug for COVID back in 2020, and several companies do versions of it. What is different is the ambition to score everything against everything and give the answers away, aimed squarely at rare conditions no drugmaker has reason to chase. The proof that it can matter is one peer-reviewed case: a man with idiopathic multicentric Castleman's disease, in hospice after every standard treatment failed, was given adalimumab — a drug for arthritis and Crohn's the model ranked at the top — and reached about two years of remission. It is the institutional version of what Fajgenbaum once did for himself, run across 18,000 diseases at once.