Graph Neural Network Analysis Of Dynamic Temporal Relationships Between Chromatin Accessibility And RNA Expression In The Developing Mouse Heart

Mammalian development is a dynamic process that transforms the epigenetic and transcriptomic landscapes of cells as they differentiate into various cell types and tissues. In vertebrates, the heart is the first functioning organ of the embryo, but its development has yet to be fully understood. Using simultaneous scRNA-seq and scATAQ-seq data from three major timepoints (E8.5, E10.5, E16.5) of murine development, we develop a novel approach to analyze the temporal relationship between chromatin accessibility and gene expression across development using RNA velocity and interpretable graph neural network (GNN) models. RNA velocity is a powerful splicing kinetics tool for uncovering directed transcriptional dynamics in complex biological systems, and it infers the direction and rate of change of gene expression in single cells over time by comparing the ratio of unspliced to spliced mRNA reads between adjacent cells. We represent the vectors identified through RNA velocity as a directed graph, thereby transforming the single-cell data with the additional simultaneous ATAC-seq information into an input for a GNN. The GNN then learns the temporal epigenetic and transcriptomic dynamics from the node feature information, ie. scATAC-seq, with the graph structure, ie. RNA velocity vectors, to predict the developmental trajectories of cells. After training of the GNN model, we implement an additional GNN explainer model to analyze it and identify subgraphs, ie. networks of DNA regions, that maximize the mutual information associated with its trajectory predictions. By interpreting the contributions of ATAC peaks in the GNN model, we identify novel regulatory dynamics and trends involved in the separation of distinct cardiac lineages in the developing mouse heart. Ultimately, we provide new insight for future investigation into the temporal relationship between chromatin accessibility and gene expression levels..