CAREER: Statistical approaches and computational tools for analyzing spatially-resolved single-cell transcriptomics data
Both healthy and diseased tissues are comprised of a multitude of interacting cells representing many different cell-types and cell-states. Rapid advances in sequencing and imaging technologies are making it possible to profile these differences for hundreds to thousands of genes in hundreds to thousands of individual cells and small groups of cells in a spatially resolved manner. However, statistical methods and computational tools are still needed to model and analyze these high-dimensional spatially resolved measurements in order to extract relevant biological insights. This project will develop statistical approaches for analyzing these spatially resolved transcriptomics datasets in order to characterize spatial gene expression patterns and associate them with underlying cellular phenotypes and functions, particularly within tissues. Such characterization will help provide insights into how spatial context and organization may impact cellular phenotype and function. The developed approaches will also be made available to the scientific community as open-source software and be applicable to other biological contexts. Finally, instructional material including case study tutorials developed through this project will be used in a hands-on-learning course to engage young girls in computer science.
On a technical level, to analyze spatially resolved transcriptomics data of tissues, this project will use a density agnostic encoding of spatial positional information using Voronoi tessellation to accommodate variations in cell density common to tissues. Spatial autocorrelation and cross-correlation analyses will be used to identify genes with significant spatial patterns that may further be indicative of cell-cell communication. Convolutional neural network-based approaches will be applied to integrate the subcellular spatial organization of mRNAs in cell segmentation within tissues and be combined with previous RNA velocity models in order to infer temporal dynamics and delineate potential cellular migration within tissues. To interrogate how gene expression variation may be associated with cellular phenotype, neural networks will also be applied to predict molecular features directly from histological staining images of tissues. Finally, a client-side web application will be developed to enable such exploration and analysis of spatially resolved transcriptomics data within the browser. For more information regarding the ongoing results of this project, please see: https://jef.works/NSF_CAREER
For more information, please see: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2047611