Abstract 6244: MultiTME: A Bayesian method for integration and analysis of multimodal, multi-patient spatial profiling data

Abstract The role of the tumor microenvironment (TME) in cancer evolution and progression is complex and multi-faceted. Multiple cellular phenotypes and patterns of cell-cell communication have been implicated as drivers of drug resistance and metastasis. The spatial architecture of the TME is emerging as particularly important in characterizing the trajectory of tumor growth and response to therapy. Accordingly, there has been a rapid rise in technologies to spatially profile the TME. Each technology focuses on a different modality (e. g. transcriptomics, proteomics, metabolomics) and operates at a different granularity, from coarse-grained tissue spots aggregating dozens of cells to fine-grained subcellular resolution. There is currently no computational method capable of drawing unified inferences across modalities. We introduce MultiTME, a single, holistic generative model for jointly modeling multimodal data across multiple samples in a cohort. MultiTME addresses these challenges through a two-dimensional integration of both samples and multimodal data. MultiTME incorporates a cohort of tissue samples, enabling the joint modeling of different samples within a single analytical framework. This integration allows for a comprehensive analysis of the spatial distribution of cell types across multiple samples, enhancing the understanding of TME variability and the identification of robust biomarkers. MultiTME also integrates multimodal data with varying resolutions, such as Spatial Transcriptomics (ST), single-cell RNA sequencing (scRNA-seq), and Imaging Mass Cytometry (IMC), and provides a multi-layered understanding of the TME. This integration enables the modeling of TME at a finer scale and the discovery of cell types and subtypes that are not distinguishable using single-modal approaches. It also provides concurrent expression profiles across different modalities, leading to the identification of novel biomarkers. We demonstrate MultiTME's efficiency and scalability with a dataset of coupled ST, scRNA-seq, and IMC from a cohort of 8 lung cancer patients. MultiTME's capability to handle both intra-cohort and inter-modality variability offers a panoramic view of TME and helps identify biomarkers and therapeutic targets for cancer screening and treatment. MultiTME is open source. Citation Format: Haoran Zhang, Jeff Quinn, Liron Yoffe, Vivek Mittal, Wesley Tansey. MultiTME: A Bayesian method for integration and analysis of multimodal, multi-patient spatial profiling data abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts) ; 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84 (6Suppl): Abstract nr 6244.