There is no such thing as a good or bad stem cell model; only models we understand and ones we do not. That was the idea that stood out to me in the 2016 work of Gioele La Manno and colleagues. Stem cell cultures hold immense promise, particularly for the treatment of neurodegenerative diseases such as Parkinson’s disease, in which replacing lost dopaminergic neurons could transform lives. However, ahead of any clinical application, we must answer a basic question: what exactly are we growing in the culture dish? To find out, the authors asked the cells themselves.

With this high-resolution reference in hand, the authors turned to stem cell models. Did these lab-grown neurons truly resemble those found in the brain? To test this, they trained a machine learning model based on logistic regression, which produced a probability map for each in vitro cell, showing how closely it matched each known midbrain cell type. Importantly, the model did not simply state whether a cell was dopaminergic, but quantified how closely it resembled the dopaminergic neurons present in the developing midbrain.