Malaria remains one of the world’s deadliest infectious diseases. The World Health Organisation counted roughly 282 million cases and 610,000 deaths in 2024. The burden falls hardest on the countries least able to absorb it. And the parasite keeps fighting back. Resistance to frontline artemisinin-based combinations now spreads through regions that once relied on them. The pipeline needs new drugs that work in new ways. It also needs them at a price the hardest-hit places can afford.
Ganaplacide is one of the most promising candidates in that pipeline. But nobody had found an affordable way to make it. This is the story of how we changed that. We used our SmartChemistry® platform to design a synthetic route no human chemist had proposed.
A drug that works, if we can make it cheaply
Ganaplacide (also known as KAF156) belongs to a class of molecules called imidazolopiperazines. It emerged from a collaboration between the Novartis Institute for Tropical Diseases, the Genomics Institute of the Novartis Research Foundation and the Swiss Tropical and Public Health Institute. Medicines for Malaria Venture and the Wellcome Trust backed the work. Its mechanism is what makes it exciting. It attacks the malaria parasite at multiple stages of its life cycle. It also stays potent against strains that already resist existing drugs.
Manufacturing has always been the catch. The original medicinal-chemistry synthesis runs to nine steps. It leans heavily on palladium-catalysed reactions, which cost a lot and leave metal impurities that chemists then have to clean up. A later version tightened the sequence but still needed costly reagents. By our estimates, none of the published routes brought the cost down far enough to matter for the communities that need the drug most.
That gap defines exactly the kind of problem we built our platform to attack: a drug that works, but that we can’t yet make affordably.
Asking the AI for a route nobody had tried
Retrosynthesis is the art of working backwards. You start with the target molecule and ask what simpler pieces could have built it. You repeat that step by step until you reach cheap, readily available starting materials. Experienced chemists do this very well. But they tend to travel well-worn paths. The interesting, cost-saving routes often hide just off the main road.
Our SmartChemistry® engine learns general transformations from enormous databases of known reactions. It then searches for ways to take a target molecule apart that a person might never try. For ganaplacide, it suggested something genuinely new. It proposed building the molecule’s central imidazole ring by joining an activated amide to a small building block called an α-aminonitrile. That single strategic move assembles the core ring system in one operation. The expert chemists who had worked on ganaplacide for years had never used this disconnection.
Creativity on its own isn’t enough, though. An AI that proposes wild, unworkable chemistry wastes everyone’s time. So we had to solve a harder problem. We had to tell promising ideas apart from fantasies.
The Synthetic Confidence Score: creativity with a seatbelt
Our answer is a metric we call the Synthetic Confidence Score. For any reaction the AI dreams up, the score measures how closely it resembles reactions that chemists have actually done and documented. It runs from 0 (no precedent) to 1 (a near-exact match). We discard anything below 0.2 automatically as too speculative. A chemist reviews anything between 0.2 and 0.4 by hand. Anything above 0.4 is confident enough to pursue.
The score earned its keep on this project. We needed to activate part of the molecule for the key ring-forming step, and we had three chemical options. The Synthetic Confidence Score ranked one of them, a methyl imidothioate, well above the others. The lab confirmed it. The top-scoring option delivered the ring cleanly. A lower-scoring alternative refused to cooperate, no matter how we pushed the conditions. The AI’s confidence and the experimental reality lined up.
The result: a route that’s dramatically cheaper
Our team then built the full route in the lab. It starts from a cheap, commercially available building block. It reaches ganaplacide through a compact sequence, with the AI-designed core assembly at its heart. We chose every key step to be practical at scale. Cost, availability, number of suppliers and safety all guided those choices, not just whether a step worked once in a flask.
The headline number is the one that matters for patients. Our route’s starting materials cost more than three times less than the original synthesis. They cost around four times less than the most recent published alternative. That gap marks the difference between a promising molecule and a deployable medicine.
Where this goes next
The Gates Foundation funded this work, and it points to something bigger than a single drug. We now use our Bayesian optimisation module to squeeze the cost down further. We are also building toward automated, robot-assisted validation so we can move faster. The same approach applies to other molecules that matter for human health: AI proposes creative disconnections, a confidence score keeps them honest, and expert chemists make the final call.
The promise here isn’t that AI replaces the chemist. It’s that AI can hand the chemist ideas they’d never find alone. Cheap medicine can be the result.
This work is described in a preprint posted to ChemRxiv in July 2026. As a preprint, it has not yet undergone peer review, and the data remain preliminary.
The full article can be seen here: https://chemrxiv.org/doi/abs/10.26434/chemrxiv.15006146/v1