PROJECT SUMMARY A decade ago, with the advent of next-generation sequencing of the human pathogen Mycoplasma genitalium, Karr et al. reported the first whole-cell model that synthesizes diverse mathematical approaches to predict a broad spectrum of biological processes. Given the recent advancements in single-cell and spatial genomics, along with the amassed cell atlas of embryogenesis, the creation of in silico models for entire mammalian embryogenesis—a long-sought goal in computational biology—seems attainable. Nevertheless, two pivotal gaps remain: (1) To capture the intricate and multi-faceted nature of embryogenesis, a cost-effective technology is requisite—one capable of profiling entire embryos at a single-cell level with high temporal resolution in 3D space. (2) To build the in silico model from the massive, high-dimensional datasets, we require powerful machine learning techniques adept at directly learning complex data-driven models and at making non-trivial predictions. In this proposal, I aim to construct the first-ever foundational in silico model of whole-embryo mouse embryogenesis. To begin, I will utilize Ultima's innovative and cost-efficient “mostly natural sequencing-by- synthesis” chemistry, combined with its ultra-high field of view wafer disc platform, to establish a large-scale 3D multi-omics cell atlas of mouse embryogenesis from E6.5 to E16.5. This will involve one-day intervals and incorporate a total of 50 million cells. The versatility of Ultima’s UG100 platform allows us to couple it with RNA metabolic labeling, CRISPR-Cas9 based lineage tracing, and multi-omics, thereby producing a comprehensive, high-definition, 3D cell atlas of mouse embryogenesis. Subsequently, I plan to devise sophisticated temporal modeling techniques for learning multi-scale, multi-modal RNA velocity vector fields. Focusing on the spatial aspect, I will devise a RNA signal-based segmentation technique for single-cell resolved spatial transcriptomics. Computer vision methods, such as the Gaussian process, will be utilized to align serial 2D slices to reconstruct the 3D embryos. To marry both temporal and spatial data dimensions, we will augment our RNA velocity vector field model to encompass data-driven PDE (partial differential equations) models. Preliminary findings suggest our model can accurately simulate the entire C. elegans embryogenesis starting from a single zygote, accounting for protein expression, cell migration, and cell fate dynamics. In parallel, to harness existing vast datasets, we'll integrate our PDE-like model with the Generative Pre-trained Transformer (as used in ChatGPT). This integration will equip our foundational model to seamlessly manage spatial, temporal, and multi-omics data. Prioritizing interpretability and predictability, we will leverage differential geometry analysis as done in my previous Dynamo framework. By merging cutting-edge technology with computational innovation, this project seeks to bridge critical gaps in our understanding of embryogenesis, enabling a first-ever in silico model of mouse embryogenesis that has the potential to revolutionize the study of developmental biology, disease mechanisms, and therapeutic interventions.

PROJECT NARRATIVE Capitalizing on recent technological strides and the amassed big data, we can now envision in silico models of mammalian organogenesis at the embryo level, offering profound predictive insights into human development and relevant diseases. This proposal seeks to pioneer such a venture by constructing the first-ever foundational in silico model of mouse embryogenesis, using Ultima's revolutionary cost-effective “mostly natural sequencing- by-synthesis” chemistry and ultra-high field of view (300, 000 mm2) spatially barcoded wafer disc platform for 3D single-cell spatial multi-omics across E6.5 to E16.5, coupled with cutting-edge machine learning methods like Fourier neural operator learning, graph neural network and the Generative Pre-trained Transformer (as employed in ChatGPT). This proposal will thus set the stage for groundbreaking predictions, including discerning the regulatory mechanisms behind, e.g., the heart's chamber formation and congenital heart disorders.