Single-Cell Perturbation Signatures Can Resolve Heterogeneous Natural Product Responses Across Cell States, Subpopulations, Pathways, and Therapeutic Sensitivities

Download PDF

Population-averaged pharmacology can compress biologically different cellular responses into the same apparent treatment effect, obscuring whether a perturbation acts broadly, selectively, incompletely, or only within particular pre-existing states. This problem is especially consequential for chemically diverse natural products, whose cellular effects cannot be assumed to be homogeneous across complex populations. This original computational framework proposes the single-cell perturbation signature as a structured pharmacological response representation that preserves baseline cellular state, perturbation-induced change, subpopulation-level response, pathway-level organization, chemical context, and uncertainty rather than reducing response to a single population statistic. The framework separates measured response from computationally inferred structure and distinguishes association with baseline state from causal state dependence. It further treats response magnitude as potentially continuous, prevents resistance from being inferred solely from persistence or incomplete response, and uses pathway decomposition as an interpretive layer rather than mechanistic proof. The proposed architecture is intended to support later modeling of compound–cell-state interactions, chemical structure, response trajectories, and breadth versus selectivity while retaining explicit boundaries around prediction, mechanism, and therapeutic interpretation. Its methodological value lies in organizing heterogeneous pharmacological information without requiring every cell to conform to a single responder class. Prospective evaluation will require perturbation-specific validation across biological contexts, assays, and chemical domains. Major limitations include confounding, sparse perturbation coverage, unstable state definitions, measurement noise, model dependence, and uncertain transportability. The contribution is therefore a testable computational framework, not a validated causal or clinical model.