TY  - JOUR
T1  - Formulation Models Should Predict Failure Modes Rather Than Only Optima When Designing Pharmaceutical Systems for Complex Natural Product Candidates
A1  - Marco De Luca
A1  - Giulia Ferraro
A1  - Antonio Russo
A1  - Elena Rossi
JF  - International Journal of Pharmaceutical And Phytopharmacological Research
JO  - Int J Pharm Phytopharmacol Res
SN  - 2250-1029
Y1  - 2026
VL  - 16
IS  - 4
DO  - 10.51847/izbWNZaggV
SP  - 99
EP  - 108
N2  - Machine-learning and data-driven formulation methods are increasingly used to predict pharmaceutical properties and identify compositions expected to maximize predefined performance objectives. For complex natural product candidates, however, a predicted optimum may be developmentally fragile if neighboring compositions, processing conditions, storage states, or physiological environments create physical instability, chemical degradation, manufacturing defects, precipitation, altered release, or loss of biological performance. This methodological perspective argues that formulation modeling should therefore represent failure explicitly rather than treating failure merely as the low-performing end of an optimization objective. We distinguish context-defined formulation failure from suboptimal performance and propose a failure-mode-first architecture in which desirable-performance targets remain analytically separate from typed physical, chemical, manufacturing, release, and biological failure targets. Failure probabilities are further distinguished from failure severity and prediction uncertainty. Stress testing is positioned as an active interrogation of predicted design space, while robust formulation regions are defined conceptually by acceptable performance maintained alongside bounded failure risk under prespecified relevant perturbations. Developmental updating should incorporate validated failures, near-failures, process changes, and distribution shifts rather than retaining only successful experiments. The framework does not establish universal failure thresholds, risk weights, or validated safe regions. Its purpose is instead to redirect formulation modeling from selecting isolated best predicted points toward identifying where performance is defensible, where models are uncertain, what can fail, and when data-driven prediction should yield to mechanistic investigation.
UR  - https://eijppr.com/article/formulation-models-should-predict-failure-modes-rather-than-only-optima-when-designing-pharmaceutica-rjbqxd9c3vjowis
ER  -