Atomwise uses deep-learning structure-based screening to predict small-molecule binding at scale, powering partnerships across pharma, crop science, and infectious disease.
Atomwise pioneered the application of convolutional neural networks to structure-based small-molecule screening. Its AtomNet technology predicts how molecules bind to target proteins, evaluating vast virtual libraries to surface candidates for testing at a fraction of the cost of physical screening. The company partners broadly — pharmaceutical companies, universities, and foundations — often with milestone-based economics that share risk. Its programs span difficult targets in oncology, infectious disease, and neuroscience, and its published results on targets lacking known binders have validated the approach where traditional screening struggles. For partners, the value is straightforward: a ranked, chemically sensible list of candidates entering the laboratory, backed by predicted binding modes and confidence estimates. Atomwise’s screening operates at scales impossible physically — billions of compounds evaluated computationally — and its model retraining loop incorporates assay results from partner programs to sharpen predictions. Its collaborative model has engaged dozens of organizations across therapeutic areas and academic institutions, broadening the range of targets and chemical hypotheses its models learn from.
Atomwise engages through partnership agreements that combine research funding with downstream milestones and royalties. The model lets organizations of any size access screening capability without large fixed investment.
Pharmaceutical companies, academic labs, and non-profit foundations pursuing small-molecule discovery — especially against targets where physical screening has stalled or proven too costly. It fits collaborations that want computational screening risk shared economically rather than paid upfront as a software license.