Current students

Understanding cancer tissue development at the cellular level requires integrative approaches that capture omics profiles, spatial organization, and the dynamic evolution of cell interactions over time. We present an Agent-Based Model (ABM) for spatiotemporal tissue growth and development, leveraging the Neural Cellular Automata (NCA) framework. More specifically, this work focus on constructing biologically meaningful cell states, which are iteratively updated by the NCA model based on neighboring spatial contexts. In order to capture the active biological pathways and gene regulations of each cell, these cell states are modeled as compact embeddings that integrate Gene Regulatory Network (GRN) activity with gene expression profiles. To achieve this, we first infer an initial GRN from single-cell RNA sequencing (scRNA-seq) data. Since GRNs inference techniques often struggle to fully capture the complex underlying biological interactions of each cell, we then refine this initial representation over time by employing advanced Graph Neural Networks (GNNs) trained on Visium HD spatial transcriptomic data from colorectal cancer organoids. The GNN model must effectively learn biologically relevant cell representations to guide decision-making of the Agent-Based Neural CA model. Additionally, in order to understand cellular responses to genetic perturbation and to capture genetic interactions involved in cancer cells, perturbation screening data, such as CRISPR-based perturbations, are incorporated to model cell state responses in a spatial and temporal cancer tissue framework.