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

Area of Science:

• Cell Biology
• Bioinformatics
• Artificial Intelligence

Background:

• Nuclear morphology provides critical phenotypic information for understanding cellular states.
• The full potential of nuclear morphology analysis remains largely untapped in biomedical research.
• Existing methods often require extensive annotations for accurate analysis.

Purpose of the Study:

• To introduce NuSPIRe, a self-supervised deep learning model for nuclear morphology analysis.
• To demonstrate NuSPIRe's capabilities in cell type identification and perturbation detection.
• To explore the integration of nuclear morphology with spatial omics data for novel biological insights.

Main Methods:

• Developed NuSPIRe, a self-supervised deep learning model utilizing DAPI-stained nuclear images.
• Pretrained NuSPIRe on a large dataset of 15.52 million cell nucleus images.
• Integrated NuSPIRe with spatial omics data to correlate nuclear structure with gene expression.

Main Results:

• NuSPIRe achieved robust performance in cell type identification and perturbation detection, even with limited annotations.
• Significant correlations were uncovered between nuclear morphology features and gene expression patterns.
• NuSPIRe demonstrated effectiveness in AI-driven experimental optimization for spatial omics.

Conclusions:

• Self-supervised learning with NuSPIRe effectively leverages nuclear morphology for biomedical analysis.
• NuSPIRe enhances data efficiency and discovery in spatial omics and molecular cell biology.
• Nuclear morphology analysis using AI offers a powerful approach to understanding cellular phenotypes and gene regulation.