Single-cell Analysis of Normal and Perturbed Dynamic and Spatial Biological Systems

National Institute Of Environmental Health Sciences

Investigators

Linked publications, trials & patents

Abstract

The aim of this project is to utilize network models as a powerful framework to investigate the complexity, spatial organization and dynamics of biological processes during development from genetic, environmental and pharmacological perturbations or exposures. In systems biology, network models provide a comprehensive understanding of the interconnections and dependencies within biological systems. They enable us to predict system behavior, build reference maps, identify potential drug targets, and discover disease biomarkers. In the last three years, we developed and published at least six major network-based algorithms: DSFMix (Anchang et al. 2022), MMQE (Atitey et al. 2023), and DEGBOE (Atitey et al. 2022) and MIBCOVIS (Atitey et al. 2024), GIBOOST (Atitey et al. 2025) and Perturb-STNet (Egbon et al. 2025). We applied these algorithms to study changes in immune, respiratory, endocrine, digestive and reproductive systems, aiming to visualize and model normal and perturbed dynamic biological processes. Algorithms like DSFMix visualizes developmental processes by analyzing single-cell data over time. It further identifies significant genes and tree motifs during development. It uses differences in distributional variations to select genes. MMQE predicts immune response dynamics based on regulatory interactions. Specifically, it explains the biological phenomenon of T cells activation followed by the proliferation of T, B and plasma B cells, as well as the antagonistic effects of IL-2 and IL-4 on lymphocyte proliferation in the presence of antigens. However, DSFMix and MMQE do not account for spatial relationships of cells during developmental process. The MIBCOVIS framework assesses and compares spatiotemporal data reduction models for optimal visualization and interpretability using a robust metric set. It provides a powerful platform for boosting the visualization and interpretability of cross tissue communication. We recently developed a new AI tool called GIBOOST (Atitey et al. 2025) which combines an Autoencoder and MIBCOVIS Bayesian framework to enhance visualization of high-dimensional data. Its performance requires selecting an appropriate performance metric of interest as well as optimizing the autoencoder for data integration. We demonstrated GIBOOST’s efficacy across multiple dynamic biological processes, including epithelial-mesenchymal transition, iPSC reprogramming, spermatogenesis, and placental development. Compared to 9 individual DRMs, GIBOOST enhances clustering sensitivity and biological relevance by about 30%, enabling more accurate interpretation of differentiation trajectories and cell-cell interactions. When applied to a large single-cell RNA-seq dataset (~500,000 cells, 28 cell types, 7 placental regions), GIBOOST uncovers novel immune-placenta interactions, providing deeper insights into cross-tissue communication during pregnancy. By improving both the visualization and interpretability of high-dimensional data, GIBOOST serves as a powerful tool for computational systems biology, enabling a more accurate exploration of complex cellular systems. More recently, we published another algorithm called Perturb-STNet (Perturbation effects with SpatioTemporal Networks). Perturb-STNet (Egbon et al. 2025) is a novel interpretable method designed to simultaneously quantify and rank spatial, and temporal deferentially expressed regulators (STDER) to measure perturbation effects on cellular behavior through time and space from single-cell data. STDER are those regulators (e.g. genes or proteins) that vary significantly across spatial locations and time and whose variability is specifically triggered by a perturbation. In this publication, we validated Perturb-STNet using synthetic data and epithelial-to-mesenchymal transition lung cancer data, showing superior performance compared to standard methods. Additionally, we applied it to CODEX single-cell imaging temporal data from a murine melanoma model to study CD8+ T-cell therapy effects, and to MERFISH spatial transcriptomics temporal data to explore inflammation and tissue repair in colitis. In melanoma, Perturb-STNet uncovered regulators like KLRG1 and CD79b, along with mediating pairs and triples (IgD-H2kb, PDL1-H2kb, NKP46-CD117, and FOXP3-CD5-CD25), revealing therapeutic strategies including checkpoint inhibition by targeting PDL1-H2kb to restore CD8+ T cell function, Treg depletion through inhibition of FOXP3-CD5-CD25 axis, and NK cell activation by enhancing NKP46-CD117 interactions. In colitis, Perturb-STNet identified key genes (Csf1r, Col6a1, Lgr4, Myc, and Fzd5) and mediator pairs (Itga5-Flnc, Cd68-Csf1r, Csf1r-Cx3cl1, and Tnfrsf1b-Bmp1) involved in immune regulation, matrix remodeling, and epithelial repair, offering potential therapeutic targets. Overall, Perturb-STNet enables robust identification of spatiotemporal regulatory networks in single-cell perturbation data across diverse disease contexts. Identifying how environmental exposures reshape spatial gene expression remains challenging due to tissue misalignment across conditions and zero-inflated gene expression in single-cell data. We recently submitted another method called Spatial-ZEDNet (Egbon et al. 2025), which is under review by Genome Biology. This framework detects spatially differentially expressed and activated genes by integrating spatial variability through a hierarchical Gaussian random field, explicitly modeling zero inflation, and aligning biological signals across conditions without requiring coordinate matching. In simulations, we showed that Spatial-ZEDNet improved spatial differential expressed genes (DEG) detection power by about 60.0% compared to existing tools like Monocle 3. Applied to colitis and Plasmodium infection spatial transcriptomics datasets, Spatial-ZEDNet identified localized expression of inflammation- and infection-associated genes, such as Mmp7 and Olr1 in colitis, and Ifitm3 and Gbp3 in Plasmodium infection, alongside spatial activation of immune pathways. Remarkably, several of the detected genes, including Tnfsf11 and Mmp8, are known genetic risk factors for inflammatory bowel disease in genome-wide association studies, linking spatial cellular responses to broader population-level disease susceptibility. Our findings highlight the importance of statistically modeling the excess zeros in spatial transcriptomics data for gene activation inferences. Spatial-ZEDNet offers a robust, interpretable approach for spatial transcriptomic data analysis and facilitates insight into exposure-induced tissue remodeling. Perturb-STNet and Spatial-ZEDNet are generic framework for spatial and spatiotemporally resolved single-cell perturbation data, and they can also accommodate non-spatially resolved data by using the high-dimensional gene expression profile to derive cell embeddings used as pseudo-spatial cell coordinates to identify differential variable genes. In collaboration with Dr. Jetten's group, we are applying Perturb-STNet and Spatial-ZEDNet on a scRNA-seq time course dataset obtained from wild-type (WT) and Glis3 knockout (KO) mice to identify the key differentiated cells and differential genes that drive the normal and deregulated β-cell developmental trajectories resulting in diabetic outcomes. This is ongoing work. In collaboration with Dr. Rodrigez’s lab , we also plan to use the Perturb-STNet and Spatial-ZEDNet algorithms to look at differences in transcriptional bursting between normal and disease states and in collaboration with Dr. Stavros lab to analyze the cross talk between cells (specifically mesenchymal and epithelia cells) over time as injury evolves from mutational processes in the lung epithelial.