DD-CC-II: Data Driven Cell-Cell Interaction Inference and Its Application to COVID-19.

Cell-cell interactions play a pivotal role in maintaining tissue homeostasis and driving disease progression. Conventional Cell-cell interactions modeling approaches depend on ligand-receptor databases, which often fail to capture context-specific or newly emerging signaling mechanisms. To address this limitation, we propose a data-driven computational framework, data-driven cell-cell interaction inference (DD-CC-II), which employs a graph-based model using eigen-cells to represent cell groups. DD-CC-II uses eigen-cells (i.e., functional module within the cell population) to characterize cell groups and construct correlation coefficient networks to model between-group associations. Correlation coefficient networks between eigen-cells are constructed, and their statistical significance is evaluated via over-representation analysis and hypergeometric testing. Monte Carlo simulations demonstrate that DD-CC-II achieves superior performance in inferring CCIs compared with ligand-receptor-based methods. The application to whole-blood RNA-seq data from the Japan COVID-19 Task Force revealed severity stage-specific interaction patterns. Markers such as FOS, CXCL8, and HLA-A were associated with high severity, whereas IL1B, CD3D, and CCL5 were related to low severity. The systemic lupus erythematosus pathway emerged as a potential immune mechanism underlying disease severity. Overall, DD-CC-II provides a data-centric approach for mapping the cellular communication landscape, facilitating a better understanding of disease progression at the intercellular level.