Abstract 5001: Multimodal topic modeling for single-cell multiomics

Abstract Single-cell (sc) multiomics assays simultaneously measure transcriptomic, epigenomic, and proteomic modalities in single cells. These multimodal assays enable deeper characterization of cells compared to single-omic assays by offering a more comprehensive measurement of the cell state. Additionally, they can identify cross modality associations, such as promoter-enhancer linkages. Recently, numerous computational methods for sc-multiomics data have been released; however, there is a dearth of interpretable methods While matrix factorization and deep learning approaches, such as MOFA+ and scVI, are effective at cell type clustering; however, these methods are difficult to interpret, masking the underlying biology that defines cell states. Our solution, EpiConfig, is a multimodal topic model that can accurately cluster cells based on cell types and yields biological insights by jointly modeling transcriptomic and epigenetic states combinations. Unlike other sc-multiomics topic models, EpiConfig explicitly models cross modality associations. Furthermore, we describe important cell filtering and subsetting approaches to speed up and improve the convergence of topic model. Benchmarking EpiConfig on publicly available scRNA+ATAC datasets to healthy and cancerous cell populations, we show that EpiConfig outperforms competing methods, such as Cobolt and MOFA. Furthermore, EpiConfig cross modality associations cell type capture 3D genome interactions. Modeling sc-multiomics data from lymphoma, EpiConfig identifies topics for cancerous cells. Coupled with our model interpretation R Shiny app, EpiConfig allows for the interpretation and identification of associations between transcriptomic and epigenomic features that define cell types and states. Citation Format: Cigdem Ak, Nicole Szczepanski, Aaron Doe, Gurkan Yardimci. Multimodal topic modeling for single-cell multiomics [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5001.