Identification of Key Behavioral Features of Cells Based on Relational Network Analysis and Its Application in Deciphering Macrophage Calcium Signaling

Abstract

High-dimensional morphodynamic data acquired through intravital microscopy contain rich biological information; however, feature redundancy and noise interference substantially limit the precise dissection of cellular functional heterogeneity. This study introduces a systematic feature‑engineering framework structured around “discriminative power evaluation – relational network construction – functional module screening” to automatically identify key behavioral descriptors from high‑dimensional parameter sets. The approach first combines the likelihood‑ratio test (LRT) and the adjusted Rand index (ARI) to quantify the discriminative efficacy of each parameter. It then integrates Pearson correlation and distance correlation to build a weighted parameter‑association network that captures both linear and nonlinear dependencies. Finally, the Leiden community‑detection algorithm is employed to identify co‑varying functional modules, from which representative key features are selected based on dual criteria: topological centrality and discriminative power.

Applied to four independent intravital imaging datasets of hepatic parenchymal macrophages (Kupffer cells), the framework robustly identifies key morphodynamic features that are strongly associated with cellular calcium‑signal intensity, while exhibiting high consistency and generalizability across experimental batches. This work not only elucidates intrinsic “morphology–function” coupling principles in macrophages, but also provides a reliable methodological tool for the unbiased quantification of immune‑cell heterogeneity in complex microenvironments.