Methods to build and annotate tissue atlases using spatial genomic data - Project Summary/Abstract Single-cell spatial technologies hold the promise of understanding spatially-resolved cell state and discovering how cells interact locally to create complex tissues and organisms. In contrast to bulk sequencing or disassoci- ated cells, with spatial genomic data we may catalog the diversity of cells and cellular environments in healthy and diseased tissues, characterize cell types and healthy variation in cell state and cellular collectives, and study the effects of local environment on cell state. However, spatial genomics technologies also pose important challenges to analytic methods, specifically, they have small fields-of-view, they are hard to normalize and align across samples, and there are limited genomic markers assayed. This proposal aims to construct queryable, multiscale, reference-free, annotated 3D tissue atlases to study how cells organize into tissues using 2D slices from spatial genomics technologies such as MERFISH, Slide-seqV2, MIBI, or even histology images. These 3D tissue atlases will allow us to align spatial genomics data and disassociated single-cell data to a 3D coordinate system. The atlases may be queried at specific 3D coordinates, within regions of interest, or for specific cell types for expected values of genomic markers. The atlases will be annotated with landmarks noting regions of interest, with genomic markers to identify those regions. Our objectives are to develop methods that build 3D tissue atlases from 2D spatial genomics samples with-out reference coordinate systems, to validate these methods by building and annotating 3D atlases for mouse heart, mouse brain, and human glioblastoma tumors using existing data, and, to identify the local influence of each aligned cell in the atlases. We will rely heavily on our expertise in Gaussian processes to infer smooth functions across spatial coordinates. The resulting atlases are able to capture cell-type specific relationships, spatial differential expression, regions of interest, and landmarks, allow integration of multimodal data and robust imputation of missing data, and are flexible with respect to biological scale, resolution, and field-of-view. To accomplish these goals, in aim one, we will develop methods to build and annotate referencefree, queryable, multiscale 3D tissue atlases from 2D spatial genomics samples. In aim two, for validation, we will build atlases with existing data for developing mouse heart, adult mouse brain, and human glioblastoma tumors. We will develop an experimental design module to allow users to robustly design their own tissue atlas with min-imal spatial genomics samples, and create a software package including a graphical user interface to visualize, explore, and analyze atlases. In aim three, we will build methods to identify the regions of local influence of each cell and cell type in an atlas. All methods will be available in GitHub through a user-friendly package. The three atlases will be publicly available to allow data alignment and searches or queries. We hope to create methods that allow single labs on limited budgets to build and study 3D tissue atlases for their tissue of interest.
