Skip to main content
2024 BMES Annual Meeting Main banner
Icon Legend
This session is not in your schedule.
This session is in your schedule. Click again to remove it.

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session E

Poster A1 - SEraster: a rasterization preprocessing framework for scalable spatial omics data analysis

Saturday, October 26, 2024
10:00 AM – 11:00 AM EST
Location: Exhibit Hall E, F & G

Introduction:  Spatial omics technologies enable high-throughout in situ molecular profiling of single cells or spots within their native tissue context at varying spatial resolutions (Park, J. et al., Genome Biology 2022). However, spatial omics data analysis scales with the number of single cells or spots, requiring excessive amounts of time and memory that can exceed the capability of current computational resources and pose scalability challenges.

Materials and

Methods:  To address this challenge, we created an R package SEraster, enabling the rasterization of spatial omics datasets during preprocessing before downstream analyses. SEraster achieves this by aggregating continuous variables (e.g., gene expression or other molecular information), and categorical variables (e.g. cell-type or cluster labels) into spatial pixels, thereby reducing the number of spatial points analyzed (Aihara, G. et al., Bioinformatics 2024). By doing so, we reduced analysis runtime and computational resource requirements while maintaining high performance in spatial variable gene analysis and cell-type co-enrichment analysis when integrated with existing tools.

Results, Conclusions, and Discussions: We benchmark the performance of SEraster in identifying spatially variable genes (SVGs) focusing on a spatial transcriptomics dataset of the mouse medial preoptic region assayed by MERFISH (Moffitt, J. et al., Science 2022). We first identify SVGs at single cell resolution to identify 150 significant SVGs. We next performed 10 rotations permutations and rasterized the dataset at 3 different resolutions (Fig. 1). We computed the positive predictive value (PPV), true negative rate (TNR), and true positive rate (TPR) of SVGs at single cell resolution versus rasterized resolutions (Fig. 2). Across all three rasterized resolutions, we observe low TNR. Investigating further, we found this low TNR can be attributed to a large number of false positive SVGs. We characterized these false positives by plotting the number of cells with non-expression as well as the proportion of cells with non-expression and found all false positives have low proportion of cells with non-zero expression (Fig. 3). As such, removing genes detected in a small proportion of cells can potentially result in more accurate identification of SVGs with rasterization. We anticipate that this rasterization preprocessing framework will enable other types of analyses and ultimately deepen our understanding of complex cellular tissue dynamics in healthy and diseased settings.

Acknowledgements (Optional):