Abstract
Protocols for deriving midbrain dopaminergic (mDA) neurons for Parkinson's disease (PD) modeling and therapy remain incompletely benchmarked against in vivo references. To establish transcriptomic standards, we generated an integrated human fetal whole-brain atlas and a midbrain subatlas. Whole-brain analysis revealed strong region-specific signatures, underscoring the need for global mapping before refined midbrain annotation. We implemented this two-tier strategy, BrainSTEM (Brain Single-cell Two tiEr Mapping), to systematically reassess published single-cell datasets of human midbrain culture models. BrainSTEM confirmed the presence of bona fide midbrain cell types ("on-target"), but also revealed substantial populations aligning with nonmidbrain regions ("off-target"), inflating reported mDA yields across protocols. This unbiased framework enables rigorous evaluation of differentiation outcomes, clarifies current limitations of midbrain-directed models, and provides a foundation for refining protocols toward more faithful in vitro systems for PD research and regenerative applications.