Researchers have developed a state-of-the-art machine learning system that combines biophysics, bioinformatics, and data mining to obtain protein structures through comparative modeling called homology, even where none exist.