Pharmacology Depends on Cell State When Computational Models Rank Natural Products Across Differentiation, Metabolic, Immune, and Disease-Specific Cellular Contexts

Download PDF

Computational models commonly rank compounds against cell lines, cell types, or disease classes as though those labels defined a stable pharmacological context. Yet the biological state encountered by a compound can vary within the same nominal system because differentiation, metabolic conditions, inflammatory activation, disease programs, treatment history, and other processes alter the molecular configuration through which perturbations are interpreted. This article develops an original computational framework for ranking natural products as conditional compound–cell-state pairs rather than as context-free compounds. The analysis integrates evidence from perturbational transcriptomics, single-cell pharmacology, developmental and lineage biology, metabolic context, immune-state mapping, disease-specific response modeling, and recent representation-learning approaches. The proposed framework separates molecular representation from cellular-state representation, models their interaction explicitly, and treats uncertainty under unseen compounds, states, and experimental systems as part of the ranking problem rather than as a downstream qualification. It further distinguishes nominal cell identity, measured cell state, predicted molecular response, functional pharmacology, and translational relevance so that one layer is not mistaken for another. Natural-product ranking is therefore reframed around state-selective activity: a candidate can be prioritized because its predicted response is favorable in a defined state and comparatively less favorable, less certain, or mechanistically different elsewhere. The framework is not presented as prospectively validated or universally generalizable. Its value is methodological: it specifies what information a cell-state-aware ranking system should preserve, what evidence can train or test it, and which biological and computational failures must remain visible before translational interpretation.