TY  - JOUR
T1  - From Misassigned Structures to Misleading Models: How Structural Annotation Errors Propagate Through Natural Product Informatics and Computational Pharmacology
A1  - Ahmed Ben Ali
A1  - Samira Gharbi
A1  - Nabil Jebali
JF  - International Journal of Pharmaceutical And Phytopharmacological Research
JO  - Int J Pharm Phytopharmacol Res
SN  - 2250-1029
Y1  - 2025
VL  - 15
IS  - 4
DO  - 10.51847/IOjlLj8ndu
SP  - 85
EP  - 93
N2  - Natural‑product informatics often treats a molecular structure as a stable input, yet the structure itself may remain an evidential claim. A misassigned constitution, configuration, regiochemistry, or molecular state can therefore enter databases and computational workflows before any model is fitted, scored, or validated. This methodological framework examines structural annotation error as an upstream modeling hazard and distinguishes experimental structure assignment from database standardization, molecular‑state preparation, descriptor generation, docking, similarity analysis, and machine‑learning prediction. Evidence from natural‑product structure revision, chemical database curation, stereochemical representation, docking benchmarks, and molecular machine learning is integrated to identify where structural uncertainty can be introduced, transformed, or hidden. The article proposes a structural‑provenance framework in which computational reuse is conditioned on traceable assignment evidence, explicit representation policy, state specification, versioned correction, and model‑sensitivity analysis. The framework does not assume that every alternative tautomer, protomer, stereoisomer, or database representation is erroneous; rather, it separates genuine assignment error from legitimate molecular multiplicity and deliberate harmonization choices. It further distinguishes chemical‑identity confidence from predictive uncertainty and numerical model confidence. The principal implication is methodological: model validation cannot compensate for an unexamined molecular identity claim when the model consumes that claim as data. The proposed framework is not a validated quality standard and does not establish the prevalence or quantitative effect of misassigned natural products in current computational datasets. Prospective testing with paired corrected and uncorrected structures is required.
UR  - https://eijppr.com/article/from-misassigned-structures-to-misleading-models-how-structural-annotation-errors-propagate-through-3uox6lf4bkrnjxa
ER  -