Across one bench, a robotic arm pipettes liquid in droplets so small they would evaporate if you stared at them too long. Across another, a stack of GPUs is digesting the assay readout from yesterday's batch. The chemists watch their next set of compounds appear on a screen, designed not by intuition or by a single overworked medicinal chemist, but by an active-learning loop that has been running, more or less, since the company opened its doors.
Kimia Therapeutics is what happens when you decide drug discovery should be measured in plates per day, not papers per quarter. The company spun out of Carmot Therapeutics in 2023 with a license to Carmot's Chemotype Evolution technology - everywhere except metabolic disease, the field Carmot kept and later sold to Roche for $2.7 billion. That sale is a useful piece of trivia. It tells you the parents knew what they were doing.
The platform Kimia is building is called ATLAS. The acronym is honest about its components - AcTive Learning with Automated Synthesis - and ambitious about its scope. The idea is to map the chemical structure-to-protein-function landscape at single-atom resolution. To draw the kind of chart that, once you have it, you cannot imagine working without.
Why bother? Because traditional med-chem is slow on purpose. It is a craft. A chemist designs, makes, tests, learns - then designs again. The throughput is bounded by the speed of a careful human and the patience of a research budget. The result, repeatedly, is a few hundred compounds explored where a few billion exist.
Kimia's bet is that the bottleneck is data, and the answer is to generate that data at industrial scale. A nanoliter-scale chemistry platform gives the company access to billions of drug-like compounds. Direct-to-biology workflows skip the long purification steps and feed compounds straight into assays. CRISPR-edited cells supply the biological readouts. And every reaction, every assay, every dead end becomes a row in the dataset that trains the next model.
None of this is unique on its own. Lots of biotechs have automated chemistry. Lots of biotechs have machine-learning teams. What Kimia is trying to build is the loop - the closed circuit where chemistry and computation iterate together fast enough that the platform actually gets smarter week over week. The honest answer to whether it works is: ask again in three years.