2024/06/24 by Nathaniel Osher, Jian Kang, Osher, Nathaniel +5
Biochemistry, Genetics and Molecular Biology · Computer Science · Medicine · #AI in cancer detection #Applications (stat.AP) #FOS: Computer and information sciences #Gene expression and cancer classification #Radiomics and Machine Learning in Medical Imaging
paper · pdf · doi:10.48550/arxiv.2406.16721
openalex publication_date 2024/06/24 · openalex created_date 2024/06/26 · openalex updated_date 2026/07/28
The spatial composition and cellular heterogeneity of the tumor microenvironment plays a critical role in cancer development and progression. High-definition pathology imaging of tumor biopsies provide a high-resolution view of the spatial organization of different types of cells. This allows for systematic assessment of intra- and inter-patient spatial cellular interactions and heterogeneity by integrating accompanying patient-level genomics data. However, joint modeling across tumor biopsies presents unique challenges due to non-conformability (lack of a common spatial domain across biopsies) as well as high-dimensionality. To address this problem, we propose the Dual random effect and main effect selection model for Spatially structured regression model (DreameSpase). DreameSpase employs a Bayesian variable selection framework that facilitates the assessment of spatial heterogeneity with respect to covariates both within (through fixed effects) and between spaces (through spatial random effects) for non-conformable spatial domains. We demonstrate the efficacy of DreameSpase via simulations and integrative analyses of pathology imaging and gene expression data obtained from 335 melanoma biopsies. Our findings confirm several existing relationships, e.g. neutrophil genes being associated with both inter- and intra-patient spatial heterogeneity, as well as discovering novel associations. We also provide freely available and computationally efficient software for implementing DreameSpase.