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.
Back

Bioinformatics, Computational and Systems Biology

Bioinformatics, Computational and Systems Biology - Poster Session E

Poster L29 - Integrating Subcellular Molecular Heterogeneity through Rasterization for Enhanced Identification of Cellular Subtypes in Spatial Transcriptomics

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

Introduction:  Identifying and characterizing transcriptionally distinct cell types and states is crucial for understanding cellular functions, spatial organization, and disease mechanisms. Misregulated gene expression in certain cells causes diseases; identifying these changes helps diagnose and understand these conditions.

Current methods, such as osmFISH, Xenium, seqFISH, and MERFISH, use spatial transcriptomic by imaging to map gene expression within cells in tissues. Current approaches to identify transcriptionally distinct cell-types and cell states rely on aggregating molecular information within cells, disregarding subcellular heterogeneity. However, subcellular regions can exhibit different molecular concentrations to support specific organelle tasks, signaling pathways, and metabolic processes. These regions’ heterogeneity influences responses to stimuli and tissue interactions. Understanding these dynamics is key to comprehending disease mechanisms, particularly in cancer and neurodegenerative disorders.

Here, we introduce a novel rasterization approach to create image representations of spatial RNA distribution within cells from molecular-resolution spatial transcriptomics data.

Materials and

Methods:  We use rasterization to convert transcript spatial coordinates into pixel-based images, capturing each cell's morphology and RNA spatial heterogeneity. To evaluate this approach, we simulated spatial transcriptomics data. Briefly, we created images of three distinct cell types with three different genes represented by the colors red, green, and blue. One of these cell types was divided into two subtypes, with the same gene expression but different subcellular organization: one subtype with gene expression concentrated in the right region of the cell and the other on the left. Each cell image was represented as a 32x32 pixel array, with the presence or absence of gene expression indicated by pixel values in a black circle and manually plotted (Fig 1).
To create tensor representations of cell segmentations and corresponding RNA species distributions, we implemented a rasterization technique. The spatial coordinates of each cell’s RNA molecules were converted into pixel intensities, with each pixel’s intensity reflecting the density and count of RNA molecules in an area of a pixel grid. A standard deviation of Gaussian interpolation kernel was applied to smooth the distribution and capture the spatial heterogeneity and density information of the RNA transcript molecules. This produced tensor representations that accurately represented the spatial distribution and concentration of RNA transcripts within the cells. These rasterized cell segmentation images were then inverted and overlayed with the rasterized gene expression images. Rasterized images for all genes in the cell are combined to form a tensor.

Results, Conclusions, and Discussions: Using our simulated data, we were able to feed these images into a pre-trained CNN, VGG16 and extract 4,016 features from this dataset. We then performed a principal components analysis (PCA) on the top 10 components and a dimensionality reduction with UMAP. When plotting a 2-D representation of the dataset, the model was able to not only cluster the different cell types, but also the sub-types of the green gene expression, suggesting evidence of certain features that can be used to identify unique gene expression profiles and subcellular RNA organization (Fig 2).

We also obtained real spatial transcriptomics data of the mouse cortex assayed by osmFISH. The data comprised cell segmentation information and RNA species distributions for individual cells. Applying our rasterization technique to the osmFISH data, we were able to create tensor-based representations given standard input formats like Loom files from imaging-based spatial transcriptomics.

Overall, we anticipate this rasterization and tensor representation approach will enable the identification and characterization of additional cell states with distinct subcellular organization and open new avenues for exploring the spatial organization of genes within cells by improving cell subtype and cell state classification as well as providing a robust framework for integrating spatial transcriptomics with advanced machine learning techniques, thereby facilitating deeper insights into cellular differentiation and cell functions.

Acknowledgements (Optional):