From Ethnopharmacological Knowledge to Drug Candidates: A Decision-Intelligence Model for AI-Guided Natural Product Discovery

Download PDF

Natural products remain important sources of pharmacologically relevant chemical diversity, yet the translation of ethnopharmacological knowledge into drug-discovery candidates is constrained by heterogeneous provenance, taxonomic ambiguity, preparation variability, uncertain disease mapping, incomplete chemical resolution, and uneven experimental support. This article proposes an original decision-intelligence model for converting ethnopharmacological knowledge into transparent, revisable, and uncertainty-sensitive candidate-selection decisions supported by artificial intelligence and computational pharmacology. The model treats reported use, documented practice, historical persistence, cross-source recurrence, practitioner knowledge, and preparation context as structured discovery inputs rather than independent evidence of efficacy, safety, or mechanism. AI-guided functions—including knowledge extraction, entity resolution, molecular representation, similarity analysis, candidate identification, target inference, disease-context matching, and multimodal evidence integration—are positioned as hypothesis-generating and decision-support functions rather than validation mechanisms. Heterogeneous evidence is translated into explicit evidence states, decision variables, uncertainty classes, and conditional rules governing progression, pause, evidence acquisition, reanalysis, escalation, rejection, reprioritization, experimental validation, and translational assessment. The proposed architecture distinguishes knowledge claims, observations, curated evidence, computational inference, orthogonal support, mechanism-level validation, and translational contextualization while preserving provenance and evidence independence. Candidate prioritization is therefore framed as a context-dependent decision problem rather than a universal ranking task. The model contributes a stage-based architecture linking knowledge provenance, source and preparation resolution, phytochemical candidate generation, AI-guided identification, target and disease inference, cross-layer evidence translation, uncertainty assessment, decision gates, experimental feedback, and translational reassessment. Its principal research implication is a shift from opaque AI ranking systems toward auditable decision-intelligence platforms capable of documenting why a candidate progresses, pauses, requires additional evidence, undergoes expert adjudication, or is rejected.