MIT's CheMeleon AI predicts molecular behavior from scant data
Engineers at MIT developed CheMeleon, a machine-learning foundation model designed to predict how new chemical compounds behave without requiring extensive physical testing. Trained on one million molecules and 1,600 structural descriptors from the PubChem database, CheMeleon learns fundamental chemical rules before adapting to target applications. When fine-tuned on small experimental datasets, the model accurately forecasts complex properties such as drug solubility, toxicity, and enzyme reactivity. By reducing the number of physical samples chemists must synthesize in the lab, CheMeleon speeds up early-stage pharmaceutical discovery and materials design while cutting research costs.
MIT News ยท Nov 6, 2026
Read on The Whole Day →