Self-supervised pretraining with NuSPIRe unlocks nuclear morphology-driven insights in spatial omics

Nuclear morphology encodes rich phenotypic information critical for understanding cellular states, yet its full potential remains untapped in biomedical analysis. This study introduces NuSPIRe, a self-supervised deep learning model designed to analyze nuclear morphology using DAPI-stained images. Pretrained on 15.52 million cell nucleus images, NuSPIRe performs robustly in cell type identification and perturbation detection, even with limited annotations. Moreover, NuSPIRe integrates nuclear morphology with spatial omics data, uncovering significant correlations between cellular structure and gene expression. Notably, NuSPIRe further enables AI-driven experimental optimization for region-of-interest identification and field-of-view selection, enhancing data efficiency in spatial omics and molecular cell biology.