Perturbation-induced gene programs map the hidden regulatory architecture of complex diseases
- 1Computational and Systems Biology Program, Sloan Kettering Institute, Memorial Sloan Kettering Cancer Center, New York, NY, USA; Physiology, Biophysics and Syst
Abstract
Steady-state molecular profiles incompletely capture the regulatory programs through which genetic variation influences complex disease, particularly programs that emerge only when cells are perturbed. Here, we introduce Perturbscape, a computational framework that learns multiscale gene programs for individual perturbations in a CRISPR Perturb-seq screen using linear and nonlinear contrastive approaches alongside complementary program-learning methods, and integrates them with human genetics to prioritize disease-relevant genes, perturbations and downstream regulatory pathways. Across 15,859 perturbation-by-context profiles from 7 Perturb-seq systems and 53 complex diseases and traits (mean GWAS N=291K), perturbation-induced programs showed up to 4.3-fold higher heritability informativeness than analogous steady-state multi-tissue gene prioritization, and retained significant conditional heritability informativeness across all 34 matched cell system-trait comparisons after accounting for steady-state prioritization. Preserving perturbation-specific programs further yielded 2.5-5.3-fold higher trait relevance scores than programs derived from control cells, pooled perturbations or differential-expression summaries. In K562, Perturbscape-prioritized genes showed stronger enrichment for rare-variant burden genes than steady-state prioritization and were supported by orthogonal enhancer CRISPRi screens. In HepG2, LDL cholesterol-associated perturbation-induced programs specifically revealed statin-responsive pathways not significantly enriched by steady-state PoPS prioritization, supported by pharmacological transcriptional responses and population plasma proteomics. Perturbscape also recovered blood-related trait-conditioned trans-regulatory relationships that were specifically supported by population trans-eQTLs in IFN-gamma-stimulated monocytes (OR = 1.30, P=1.6e-05). Perturbscape further revealed developmental-stage-specific convergence between monogenic diabetes regulators and polygenic T2D risk during pancreatic islet differentiation, and neuronal-subclass-specific autism-associated regulatory programs in vivo. Together, these findings establish perturbation-induced gene programs as a complementary layer of disease-relevant regulatory information, revealing genetically informative pathways and context-specific regulatory relationships incompletely represented by steady-state molecular atlases.
Related
- Where single-cell foundation models help: a benchmark of annotation, perturbation and regulatory inference against simple baselines
- Genetic risk and inflammatory signaling converge on cell type-specific regulatory programs in type 1 diabetes
- Transcriptional regulatory programs underlying cellular and regional specialization in the late-gestation porcine brain
- A hierarchical regulatory cascade programs chlamydospore formation in the nematode-trapping fungus Arthrobotrys flagrans
- A Mechanistic Viral Host Perturbation Score (VHPS) Reveals Shared and Virus‐Specific Transcriptional Programs Across Human Respiratory Infections
BibTeX
@misc{turhan2026perturbationinduced,
title = {{Perturbation-induced gene programs map the hidden regulatory architecture of complex diseases}},
author = {B. Turhan and H. Lee and D. Liu and A. Le and A. Ghosh and D. Nandi and R. Mazumder and J. M. Engreitz and X. Jin and D. Huangfu and K. K. Dey},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.03.762080},
url = {https://www.biorxiv.org/content/10.64898/2026.11.03.762080v1},
note = {CC BY-NC},
}
RIS
TY - UNPB
TI - Perturbation-induced gene programs map the hidden regulatory architecture of complex diseases
AU - B. Turhan
AU - H. Lee
AU - D. Liu
AU - A. Le
AU - A. Ghosh
AU - D. Nandi
AU - R. Mazumder
AU - J. M. Engreitz
AU - X. Jin
AU - D. Huangfu
AU - K. K. Dey
PY - 2026
DA - 2026/11/03
PB - bioRxiv
DO - 10.64898/2026.11.03.762080
UR - https://www.biorxiv.org/content/10.64898/2026.11.03.762080v1
AB - Steady-state molecular profiles incompletely capture the regulatory programs through which genetic variation influences complex disease, particularly programs that emerge only when cells are perturbed. Here, we introduce Perturbscape, a computational framework that learns multiscale gene programs for individual perturbations in a CRISPR Perturb-seq screen using linear and nonlinear contrastive approaches alongside complementary program-learning methods, and integrates them with human genetics to prioritize disease-relevant genes, perturbations and downstream regulatory pathways. Across 15,859 perturbation-by-context profiles from 7 Perturb-seq systems and 53 complex diseases and traits (mean GWAS N=291K), perturbation-induced programs showed up to 4.3-fold higher heritability informativeness than analogous steady-state multi-tissue gene prioritization, and retained significant conditional heritability informativeness across all 34 matched cell system-trait comparisons after accounting for steady-state prioritization. Preserving perturbation-specific programs further yielded 2.5-5.3-fold higher trait relevance scores than programs derived from control cells, pooled perturbations or differential-expression summaries. In K562, Perturbscape-prioritized genes showed stronger enrichment for rare-variant burden genes than steady-state prioritization and were supported by orthogonal enhancer CRISPRi screens. In HepG2, LDL cholesterol-associated perturbation-induced programs specifically revealed statin-responsive pathways not significantly enriched by steady-state PoPS prioritization, supported by pharmacological transcriptional responses and population plasma proteomics. Perturbscape also recovered blood-related trait-conditioned trans-regulatory relationships that were specifically supported by population trans-eQTLs in IFN-gamma-stimulated monocytes (OR = 1.30, P=1.6e-05). Perturbscape further revealed developmental-stage-specific convergence between monogenic diabetes regulators and polygenic T2D risk during pancreatic islet differentiation, and neuronal-subclass-specific autism-associated regulatory programs in vivo. Together, these findings establish perturbation-induced gene programs as a complementary layer of disease-relevant regulatory information, revealing genetically informative pathways and context-specific regulatory relationships incompletely represented by steady-state molecular atlases.
ER -