An inverse mixture-design approach based on machine learning can teach computers to create mixtures from a set of target properties. Developed by KAUST, this could help find high-performance transport fuels that burn efficiently while releasing little carbon dioxide (CO2) into the atmosphere.