Beyond correlation: Causal discovery of cell program interactions in lung tissue

Cell-cell communication networks govern tissue function in health and disease, but existing computational tools rely on static ligand-receptor databases and correlation-based analyses that cannot distinguish direct causal relationships from confounding or indirect associations. We introduce Cell-CLIP (Cellular Causal Learning and Inference Pipeline), a modular framework applying Pearls causal discovery algorithms to infer directed cell program interaction networks from patient-level single-cell RNA-seq data. Cell-CLIP aggregates noisy single-cell expression into biologically interpretable cell program activity scores using JASMINE51, then applies PC (constraint-based) and GES (score-based) causal discovery algorithms to learn directed acyclic graphs representing putative causal relationships. We applied Cell-CLIP to a 176-patient lung atlas (normal, COVID-19, adenocarcinoma) focusing on 14 universal cell programs active across all phenotypes. Hybrid consensus causal discovery (PC and GES union) identified 21 directed edges, yielding a sparse network (11.5% density) dominated by macrophage program hubs. Four edges discovered by both algorithms (Intersection consensus) agree with known biological knowledge, included well-validated biological mechanisms like: CD8/NK differentiation cascades, macrophage sensing-recruitment etc. Quantitative benchmarking against a 40-edge Pearson correlation baseline demonstrated Cell-CLIPs complementary advantages: while correlation showed low specificity (28% of edges truly causal) and sensitivity (52% recall of causal edges), Cell-CLIP achieved 100% specificity by eliminating spurious associations from confounding and indirect paths. Key methodological advances include: correction for confounding by rejecting spurious correlations, detection of masked direct effects via conditional independence testing, inference of causal directionality distinguishing sequential regulatory cascades from coregulation, and generation of novel predictions for experimental validation. Cell-CLIP provides a hypothesis-generating framework for discovering candidate causal circuits in cell-cell communication that extends beyond correlation-based methods, though inferred relationships require experimental validation due to observational data limitations.