Automated assignment of cell identity from single-cell multiplexed imaging and proteomic data

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning in Biology

Background:

• Cell type assignment in highly multiplexed imaging data is challenging.
• Current methods rely on manual annotation, which is subjective and cannot identify novel cell types.

Purpose of the Study:

• To present Astir, a probabilistic model for automated cell type assignment.
• To integrate prior knowledge of marker proteins for accurate cell classification.
• To enable cell type assignment at scale and identify uncharacterized cell types.

Main Methods:

• Developed Astir, a probabilistic model utilizing deep recognition neural networks.
• Integrated prior knowledge of marker proteins for cell type assignment.
• Applied the model to large-scale suspension and imaging datasets (over 2.4 million cells).

Main Results:

• Astir enables cell type assignment at the million-cell scale.
• Demonstrated scalability and robustness to varying sample compositions.
• Provided interpretable uncertainty estimates for cell assignments.
• Successfully assigned cells in the absence of a previously annotated reference.

Conclusions:

• Astir offers a robust and scalable solution for cell type assignment in complex biological data.
• The model facilitates broad cell type classification and the identification of potential biomarkers.
• Astir enhances the analysis of multiplexed imaging data, advancing biological discovery.