Mapping human microglial morphological diversity via handcrafted and deep learning-derived image features.

Microglia regulate brain health and disease through diverse, dynamic activation states, but capturing this continuous heterogeneity at scale remains challenging. We developed an imaging and analysis framework to map activation landscapes of human iPSC-derived microglia (iMG) at single-cell resolution. High-content imaging combined a hypothesis-driven immunofluorescence (IF) panel targeting NF-κB, ASC, and CD45 with a discovery-oriented cell painting (CP) assay. Phenotypes were quantified using handcrafted and representation-learning features. To classify cells, we applied Gaussian mixture models (GMMs), enabling soft probabilistic assignments that capture transitional states. Compared with graph-based methods such as Leiden, GMMs achieved similar performance while providing more interpretable descriptions of microglial heterogeneity. Deep-learning features from the targeted IF panel were most informative, yielding high classification accuracy and strong correlation with biological readouts, including NLRP3 inflammasome activation. This platform offers a scalable approach to quantify microglial states and provides a scalable platform for discovering compounds that modulate microglial phenotypes.