Researchers used machine learning to identify two main types and five subgroups of Parkinson's disease, explaining treatment variation.