Inferring Personalized Cell-Cell Communication Networks in Colorectal Cancer with Individualized Causal Discovery.

Understanding tumor heterogeneity at the resolution of individualized cell-cell communication networks (CCCNs) remains a major computational challenge in precision oncology. Existing inference methods largely rely on population-level correlation and thus fail to capture patient-specific signaling patterns across diverse cell types. To address this limitation, we developed an integrative computational framework combining the nested hierarchical Dirichlet process (nHDP) model for identifying hierarchically structured gene expression modules, with instance-specific Greedy Fast Causal Inference (iGFCI) for inferring individualized CCCNs (iCCCNs) in colorectal cancer (CRC). Applied to large-scale single-cell RNA-seq data from over 625,000 cells, our model successfully decomposed complex gene expression modules GEMs, potentially representing the cell lineage and cellular signaling states, and uncovered iCCCNs across detailed cell subtypes. We further used TCGA bulk RNA-seq data with survival data to validate the clinical relevance of these individualized gene expression module causal interactions, demonstrating their potential as robust prognostic signatures in CRC. Finally, we used principled causal inference methods to search for ligand-receptor pairs that mediate cell-cell communication. This framework enables mechanistic insights into immune evasion. Our computational method represents a significant advance toward realizing personalized oncology, enabling precise patient stratification and identification of actionable biomarkers for improved therapeutic targeting in cancers.