Deconvolution of cell types and states in spatial multiomics utilizing TACIT

Area of Science:

• Spatial biology
• Cellular and molecular biology
• Computational biology

Background:

• Accurate cell type and state identification is crucial but challenging in spatial biology.
• Existing methods struggle with data variability and generalization across health and disease contexts.
• Deep learning approaches often require extensive training data and lack adaptability.

Purpose of the Study:

• To develop an unsupervised algorithm for robust cell annotation in spatial biology.
• To overcome limitations of existing methods in accuracy, scalability, and data requirements.
• To enable novel discoveries in complex biological systems and diseases.

Main Methods:

• Developed TACIT (unsupervised algorithm for cell annotation using predefined signatures).
• Employed unbiased thresholding for cell identification and marker-focused analysis for ambiguous cells.
• Validated on five large-scale datasets across brain, intestine, and gland niches.

Main Results:

• TACIT demonstrated superior accuracy and scalability compared to existing unsupervised methods.
• Identified novel cellular phenotypes in inflammatory gland diseases.
• Revealed under- and overrepresented immune cell types/states using multiomic spatial data.

Conclusions:

• TACIT provides a powerful, data-efficient tool for cell annotation in spatial biology.
• Unsupervised cell annotation advances understanding of cellular heterogeneity in health and disease.
• Multimodal spatial analysis is essential for clinical translation of spatial biology findings.