Abstract 1203: Mushroom: a tool for identification of 3D cellular neighborhoods in multi-modal spatial datasets

Abstract A comprehensive understanding of the spatial localization of cellular processes is necessary to fully describe tissue biology. Numerous techniques exist to identify cellular neighborhoods in spatial datasets, however, these techniques are limited in that they either only capture two-dimensional phenomena - and can't describe the full, 3-dimensional tumor volume, or they are restricted to one spatial technology (i.e. IF, 10X Visium, Xenium, MERFISH, etc.). To address this, we developed Mushroom, a tool for identification of three-dimensional cellular neighborhoods in serial-sectioned multi-modal datasets. Mushroom’s methodology consists of two main steps: 1) registration of multi-modal technologies taken in serial sections (such as multiplex imaging, H&E, or spatial transcriptomics), and 2) identification of 3-dimensional cellular neighborhoods comprising the tissue volume that has been serially sectioned. Keypoint registration is performed with BigWarp, after which the resulting dense displacement field (DDF) is used to register neighboring serial sections - including those from different data modalities. We then implement a neural network to identify three dimensional cellular neighborhoods. The network is a variational autoencoder (VAE) with a vision transformer (VIT) backbone that has been adapted to work with multiple spatial data modalities. Features extracted from the VIT backbone are then clustered to identify cellular neighborhoods unique to each input data type. With 3D neighborhoods in hand, biological questions not answerable in traditional 2D approaches can be investigated, such as z-dimensional changes in morphology and cellular interactions. Further, seamless characterization of cellular neighborhoods by simultaneous technologies at multiple levels of spatial and genomic resolution becomes possible. Citation Format: Erik Storrs, Siqi Chen, Chia-Kuei Mo, Andrew Houston, Andrew Shinkle, Austin Southard-Smith, Faria Simin, Andre Targino, Xiang Li, Ateih Abedin, Reyka Jayasinghe, Alla Karpova, Jingxian Liu, John Herndon, David Fenyo, Feng Chen, Ryan Fields, Tao Ju, Ben Raphael, Li Ding. Mushroom: a tool for identification of 3D cellular neighborhoods in multi-modal spatial datasets [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(6_Suppl):Abstract nr 1203.