CSCD: Towards Spatially Resolving the Heterogeneous Landscape of MxIF Oncology Data

Contrasting spatial co-location pattern discovery aims to find subsets of spatial features whose prevalences are substantially different in two spatial domains. This problem is important for generating hypotheses in many spatial applications, including oncology, regional economics, ecology, and epidemiology. In oncology, for example, this problem is important in developing immune-checkpoint inhibitor therapy for cancer treatment. This problem is challenging due to the large number of potential patterns that are exponentially related to the number of input spatial features. Traditional methods of co-location pattern detection require multiple runs, making computationally expensive and do not scale to large datasets. To address these limitations, we propose a Contrasting Spatial Co-location Discovery (CSCD) framework and contribute two filter-refine algorithms that exploit a novel interest measure; the participation index distribution difference (PIDD). Experiments on multiple cancer datasets (e.g., MxIF) show that the proposed algorithm yields substantial computational time savings compared with a baseline algorithm. A real-world case study demonstrates that the proposed work discovers patterns that are missed by the related work and have the potential to inspire new scientific discovery.