Learning Directed Acyclic Graphs for Ligands and Receptors Based on Spatially Resolved Transcriptomic Analysis of Ovarian Cancer

• Activity Number: 308 - Statistical Methods for Studying Spatial Transcriptomics, Tissue Heterogeneity, and Pleiotropy
• Type: Topic-Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): SHRABANTI CHOWDHURY* and Sammy Ferri-Borgogno and Peng Yang and Jie Peng and Wenyi Wang and Samuel Mok and Pei Wang
• Companies: Icahn school of Medicine at Mount Sinai and MD Anderson Cancer Center and University of Texas MD Anderson Cancer Center and University of California Davis and University of Texas MD Anderson Cancer Center and MD Anderson Cancer Center and Icahn school of Medicine at Mount Sinai
• Keywords: Causal association; Bootstrap aggregation; Prior; Hill climbing; zero-inflation

Abstract

In immune activation/suppression mechanisms, a major type of cell-cell communication between tumor cells and immune/stromal cells in tumor micro-environments is signaled through interactions between ligands of one cell and cognate receptors of a neighboring cell. Recently emerged technology of spatial transcriptomic (ST) profiling provides us the unprecedented opportunity to systematically study these important ligand-receptor interactions in tumor micro environment. In this work, we propose a new method, DagBagST, to construct ligand-receptor DAGs based on ST data. In DagBagST, we first introduce binary variables to indicate the on/off status of ligands/receptors which helps to account for zero inflated distribution in the ST data. We then utilize a modified hill-climbing algorithm to build DAGs for both continuous and binary nodes. In addition, we make use of the known ligand-receptor regulation information from relevant database to constrain the searching space when constructing DAGs. We illustrate the performance of DagBagST through simulation studies and a ST data set of an ovarian cancer study.