Graph of graphs analysis for multiplexed data with application to imaging mass cytometry.

Area of Science:

• Computational Biology
• Biotechnology
• Medical Imaging

Background:

• Imaging Mass Cytometry (IMC) enables high-resolution, multiplexed protein analysis in intact tissues.
• IMC data presents complex, high-dimensional spatial information across multiple regions of interest (ROIs).
• Extracting meaningful biological insights from complex IMC datasets remains a challenge.

Purpose of the Study:

• To develop a novel graph-based analytical framework for IMC data.
• To create a method for integrating information from multiple ROIs within IMC assays.
• To improve the prediction of treatment response using IMC data.

Main Methods:

• A two-step graph-based approach was developed to analyze IMC data.
• Graphs were constructed for each ROI to represent feature interactions, followed by a graph of ROIs.
• Nonlinear dimension reduction was applied to the ROI graph to derive phenotypic representations.

Main Results:

• The proposed method effectively organizes ROIs based on their intrinsic geometric representation.
• The approach demonstrated superior distinction between modalities compared to traditional estimators in specific distributions.
• Applied to lung cancer data, the method achieved 97.3% accuracy in predicting sensitivity to PD-1 axis blockers.

Conclusions:

• The graph of graphs approach provides an informative representation of IMC data by integrating ROI-level and feature-level information.
• This method enhances the association between spatial data and phenotypic states of the entire image.
• The framework shows significant potential for advancing precision medicine through spatial biology analysis.