New AI Model Rapidly Predicts Drug Binding Affinity With High Accuracy

Study in a Sentence: A new multi-component, transformer-based, open-source AI model, Boltz-2, predicts protein–drug binding affinity with state-of-the-art accuracy at 1000 times the speed, unlocking new capabilities to understand how drugs behave, which is critical to predicting drug toxicity.

Healthy for Humans: Researchers introduced Boltz-2, an AI tool similar to AlphaFold, that accurately predicts protein–ligand structures and their binding affinities. Boltz-2 achieves prediction accuracy approaching that of highly intensive physics simulation methods, while being orders of magnitude faster. This breakthrough opens up new and exciting possibilities, like screening drugs against the entire human proteome to predict efficacy and toxicity—without animal testing.

Redefining Research: Boltz-2 goes beyond structure prediction and enables fast, cost-effective evaluation of how well drugs interact with both intended targets and potential off-targets. By identifying which targets and how well a drug binds to them, researchers can better predict the physiological consequences of the drug. AI-based drug testing could lead to more accurate predictions of a drug’s effectiveness and toxicity in humans than experiments on animals.