Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.64898/2025.12.01.691666

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

Al-Tal, M. K. F.·Liu, X.·Zhou, J.
View original source ↗Export metadata ↓

Abstract

Cell-cell communication networks govern tissue function in health and disease, yet existing computational tools rely on static ligand-receptor databases or correlation-based analyses that cannot distinguish direct causal relationships from confounded or indirect associations. We introduce Cell-CLIP (Cellular Causal Learning and Inference Pipeline), a modular framework that combines (i) mask-aware patient-level JASMINE activity scoring with low-percentile imputation in raw activity space, (ii) five complementary causal-discovery algorithms -- PC, GES, FCI, DirectLiNGAM, and GRaSP -- each fed an algorithm-matched per-column transform of the same masked JASMINE matrix, (iii) direction-preserving bootstrap aggregation with a weighted two-phase consensus, (iv) Joint Causal Inference (JCI) for principled pooling of patient cohorts via context indicators, and (v) a residual-independence (RESIT) direction audit, all evaluated under a systematic three-by-three-by-three hyperparameter sweep. Applied to a 611-patient multi-phenotype lung atlas spanning normal tissue, COVID-19 pneumonia, and lung cancer, the cross-cohort joint-causal-inference graph constructed by stratifying patients into normal, COVID-19, and tumour contexts before pooling recovers 100% of evaluable curated ground-truth edges (8/8; Wilson 95% CI 0.68-1.00) with 75.0% direction accuracy (6/8; 95% CI 0.41-0.95) and zero forbidden-orientation violations (0/8; 95% CI 0.00-0.32) -- the only configuration in a comprehensive ablation series spanning 7 cohort/pool configurations, 27 hyperparameter settings, and 3 consensus types to simultaneously reach all three optimal Pareto corners. The single-cohort universal cross-phenotype graph (12 cross-phenotype programs over the full 611-patient atlas) similarly achieves 100% evaluable recall (8/8; 95% CI 0.68-1.00) with 62.5% direction accuracy (5/8; 95% CI 0.31-0.86) and zero forbidden violations (0/3; 95% CI 0.00-0.56) under the same systematic hyperparameter sweep with low-percentile imputation, exceeding an earlier zero-imputation baseline configuration of the pipeline on every reported metric (Table 6, Table 8); given small ground-truth denominators (n_eval = 8), these point-estimate improvements lie within the Wilson 95% intervals of the baseline and should be read as direction-of-effect rather than as confirmatory comparisons. We further show that single-cohort causal inference on the disease-extended pooled cohorts (which apply the COVID- and cancer-extended program sets to all 611 patients) recovers strong skeleton recall (35.0%-55.2% evaluable) but suffers a structural direction collapse (down to 14.3% direction accuracy and 60% forbidden rate at union) caused by within-cohort averaging over heterogeneous normal-plus-disease patient populations; the patient-stratified joint-causal-inference pool dissolves this collapse without altering the underlying algorithms. Cell-CLIP provides a hypothesis-generating framework for inferring directed cell-program networks from patient-level scRNA-seq atlases, with all reported relationships requiring experimental validation due to observational data limitations. O_TBL View this table: org.highwire.dtl.DTLVardef@1b18810org.highwire.dtl.DTLVardef@5c027eorg.highwire.dtl.DTLVardef@a7b806org.highwire.dtl.DTLVardef@122adf2org.highwire.dtl.DTLVardef@1ca0a39_HPS_FORMAT_FIGEXP M_TBL O_FLOATNOTable 6C_FLOATNO O_TABLECAPTIONHeadline validation metrics across reporting cohorts, with Wilson 95% confidence intervals on every binomial rate. C_TABLECAPTION C_TBL O_TBL View this table: org.highwire.dtl.DTLVardef@1653cb1org.highwire.dtl.DTLVardef@173c6baorg.highwire.dtl.DTLVardef@1fc115forg.highwire.dtl.DTLVardef@1d87e39org.highwire.dtl.DTLVardef@46d53b_HPS_FORMAT_FIGEXP M_TBL O_FLOATNOTable 8C_FLOATNO O_TABLECAPTIONAblation matrix: evaluable recall at the recall-best configuration. Variants are defined in the Sensitivity analysis section. Cells marked not run were not executed for that combination of cohort x variant; reasons are given in the table caption. C_TABLECAPTION C_TBL

Explore related subjects

systems biologyPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Al-Tal, M. K. F., Liu, X., Zhou, J.. 2025-12-03. Beyond correlation: Causal discovery of cell program interactions in lung tissue. https://doi.org/10.64898/2025.12.01.691666

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

INFORME: coupling information-theoretic experimental design with nonlinear mixed-effects modeling for efficient observation scheduling

Mathematical models of treatment response can inform individualized therapy, but their calibration often requires longitudinal measurements that are costly, burdensome, and collected on fixed schedules. Such schedules may be inefficient, over-sampling patients whose response is already well characterized while delaying informative measurements for those whose model parameters remain uncertain. We present INFORME (INFORmation-theoretic design with Mixed Effects), a framework that combines Bayesian information-theoretic experimental design with nonlinear mixed-effects modeling to adaptively select each patients next measurement time. Population and response-subgroup parameter distributions learned from an existing cohort provide informative priors, allowing candidate measurement times to be ranked by their expected reduction in patient-specific parameter uncertainty. As observations accumulate, priors can be updated to reflect the response subgroup most consistent with the patients data. We evaluate INFORME in two radiotherapy datasets: 150 synthetic tumor volume trajectories from a hybrid cellular automaton model of prostate cancer spheroids (HD1) and longitudinal tumor volumes from 39 patients with head-and-neck cancer (HD2). In HD1, population priors allowed omission of both pretreatment scans, while adaptive scheduling reduced the protocol from nine scans to three or four, with the response group identified from a single post-treatment scan on day 27. In HD2, the adaptive schedule used three scans instead of six and improved prediction by delaying the first on-treatment scan from week 1 to week 2, avoiding transient dynamics that produced false-positive and false-negative response projections. Across both datasets, the adaptive schedules used a mean of 2.7 scans in stead of seven and advanced completion of the patient-specific prediction by a mean of 15.5 days (95% CI, 6.7-24.3) relative to the equidistant protocol, while treatment duration remained unchanged. INFORME therefore reduces measurement burden and accelerates patient-specific prediction by concentrating observations at times that are most informative for model calibration.

systems biology↗
bioRxiv2026

Sobetirome, a thyroid hormone receptor beta agonist, is a potential therapeutic agent for pulmonary fibrosis

Idiopathic pulmonary fibrosis (IPF) is a progressive and fatal disease with limited treatment options. Our group previously identified the antifibrotic potential of thyroid hormone, triiodothyronine (T3); however, clinical translation of thyroid hormone therapy is limited by its systemic adverse effects. In this study, we investigate whether sobetirome, a selective and well tolerated thyroid hormone receptor beta (THRB) agonist, offers antifibrotic benefits of thyroid hormone while minimizing systemic toxicity. Our study reveals that sobetirome, administered via intraperitoneal or inhalational routes, effectively mitigates bleomycin-induced pulmonary fibrosis in mice, with no evidence of toxicity. We identified that sobetirome restores mitochondrial homeostasis via activating the THRB-PPARGC1a axis. This protects alveolar type II epithelial cells from injury-induced apoptosis while selectively inducing apoptosis and metabolic reprogramming in apoptosis resistant IPF fibroblasts. Cell-specific deletion of Ppargc1a in either alveolar epithelial cells or fibroblasts abolishes sobetirome-mediated protection, establishing PPARGC1a as an essential mediator of therapeutic response. Importantly, sobetirome reverses fibrosis-associated transcriptional programs in human IPF lung tissue, reducing expression of key fibrosis-associated genes, including collagen I alpha 1 (COL1A1), collagen III alpha 1 (COL3A1), periostin (POSTN), cathepsin K (CTSK), and Chitinase 3 Like 1 (CHI3L1), while promoting extracellular matrix remodeling, epithelial restoration, and tissue homeostasis. Collectively, our findings identify THRB activation as a novel metabolic strategy for reversing pulmonary fibrosis. Across complementary in vitro, in vivo, and human ex vivo models, sobetirome restores mitochondrial function, modulates apoptotic pathways in pathogenic cells, and promotes fibrosis resolution, highlighting its potential as a lung-targeted therapeutic approach for IPF and other fibrotic lung diseases.

systems biology↗
bioRxiv2026

Mechanistic modeling of bacterial translation initiation across growth conditions

Translation frequency in bacteria depends on how ribosomes, mRNAs, and initiation factors are allocated across growth conditions. Here, we developed a mechanistic ODE-based model of Escherichia coli translation that represents initiation, elongation, termination, and coupled auxiliary processes. Growth-dependent abundances were derived from physiological relationships and reprocessed omics data, and simulated outputs were compared with translation-frequency and active-ribosome references. The model predicts a continuous shift from complex-formation-limited toward ribosome-limited behavior as growth increases. This shift is characterized by a decline in free-ribosome abundance, whereas initiation-factor pools remain largely unbound and do not become depleted in parallel. Together with the implemented IF-dependent kinetic term, this preserved availability provides a model-internal route through which productive initiation can be maintained despite increasing ribosome utilization. Consistently, transcript-wide ribosome loading remains below its theoretical maximum, while COG-level simulations reveal distinct sector-specific translation-frequency trajectories. The study therefore provides a resource-allocation framework for interpreting how mRNA--ribosome interactions shape bacterial translation across growth conditions.

systems biology↗