AN EXPLORATION OF METHODS FOR QUANTIFYING MULTIPLE CELL TYPE INTERACTIONS

Spatial transcriptomics and multiplex immunofluorescence imaging have enabled high-resolution mapping of cellular interactions in tissue samples, providing deeper insights into thespatial organization of tumor-infiltrating lymphocytes (TILs). To compare cell interactions acrossa collection of tissue samples, careful consideration must be given to the representation of cell lo-cation data. This dissertation focuses on methods for representing multi-label point pattern dataobjects to study variation in cell interactions among breast cancer tissue samples.In the first two projects, we focus on interactions between CD8+ T cells, an important typeof cytotoxic immune cell, and tumor cells. We develop two spatial biomarkers based on the dis-tribution of nearest neighbor distances of T cells with respect to tumor cells and show that theyare prognostically significant for breast cancer recurrence outcomes. We extend these results byapplying tools from topological data analysis. Utilizing a function of nearest neighbor distances, wecan capture multi-scale spatial information in a single vector representation based on the changingtopology of a sequence of nested sublevel sets. This representation led to the discovery of additionalbiologically relevant directions in the data space.In the third project, we introduce the co-clustering matrix data object as a method for captur-ing interactions between multiple pairs of cell types represented in each point pattern. Interactionamong cells of the same type as well as interactions between each pair of cell types are measured us-ing a transformation of the Ripley’s K-function and collected in a matrix. The co-clustering matrixrepresentation provides an interpretable visualization of all single and pairwise cell-type interac-tions simultaneously. Application of this method to breast cancer tissue cell location data revealsdirections in the data space corresponding to effective immune infiltration as well as directions thatsignificantly differentiate between key breast cancer subtypes.