Enhancing spatial genomic analysis of tumor heterogeneity: Addressing variability, batch effects, and confounding factors with extended GeoMx package
Presenting author: Kun Bai, Department of Biostatistics, Vanderbilt University Medical Center
Co-authored by:
- Fei Ye, Department of Biostatistics, Vanderbilt University Medical Center
- Yan Guo, Department of Public Health and Sciences, University of Miami
Abstract:
The GeoMx package facilitates advanced spatial analysis of gene expression. To comprehend gene expression in different tumor segments within a patient and across cancer sites, it's important to account for this complex variance structure while adjusting for potential confounding factors, including batch, race, and geographical region. Our goal was to expand the GeoMx package by 1) incorporating batch effect correction and normalization steps, 2) performing unsupervised clustering and principal component analysis to identify patient subgroups, and 3) capturing the interperson versus intraperson variability within the same segment of the tumor by modelling the hierarchical variance structure. We aimed to address questions surrounding tumor heterogeneity, including the estimation of variability among segments within tumor, potential geographical differences across tumor types, and biomarker effects. We developed multi-level mixed models to estimate different sources of variability, while accounting for batch effects and other confounding factors. Our tool leverages the GeoMx R package to enhance spatial genomic analysis by addressing the challenges arising from the spatially structured data and multiple sources of variability and bias. We have demonstrated in a real-world dataset that our tool can provide additional insights into spatially distinct biological processes that might influence disease development, diagnosis, and treatment decisions.