Spatial Deconvolution of Cell Types and Cell States at Scale Utilizing TACIT.

Area of Science:

• Spatial biology
• Computational biology
• Genomics

Background:

• Cell type and state identification is crucial but challenging in spatial biology.
• Current deep learning methods struggle with generalization due to biological variability.
• Need for robust, data-independent cell annotation methods.

Purpose of the Study:

• Develop TACIT, an unsupervised algorithm for accurate cell annotation in spatial biology.
• Overcome limitations of existing methods in accuracy and scalability.
• Enable discovery of novel cell phenotypes and states in health and disease.

Main Methods:

• Developed TACIT, an unsupervised algorithm using predefined signatures and unbiased thresholding.
• Applied TACIT to five diverse spatial biology datasets (5 million cells, 51 cell types).
• Integrated TACIT with a Shiny app for phenotype discovery and analyzed multiomic data.

Main Results:

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

Conclusions:

• TACIT provides a robust, data-independent solution for cell annotation in spatial biology.
• Unsupervised cell annotation is essential for advancing spatial biology research.
• Multimodality is key for translating spatial biology findings into clinical applications.