TY  - JOUR
T1  - Explainable and Uncertainty-Aware QSAR for Pharmaceutical Decision Making
A1  - Aisha Bello
A1  - Zainab Sule
A1  - Ibrahim Musa
A1  - Grace Adeyemi
A1  - Ahmed Youssef
JF  - International Journal of Pharmaceutical Research and Allied Sciences
JO  - Int J Pharm Res Allied Sci
SN  - 2277-3657
Y1  - 2026
VL  - 15
IS  - 3
DO  - 10.51847/wgZwhbZLOq
SP  - 74
EP  - 84
N2  - Quantitative structure–activity relationship (QSAR) models increasingly support pharmaceutical prioritization, yet predictive accuracy alone does not establish whether a model is sufficiently interpretable or reliable for a specific decision. Explainability methods may identify molecular features associated with predictions, whereas uncertainty methods estimate confidence, calibration, or susceptibility to error; these functions are related but not interchangeable. A systematic review of literature published from 2017 to September 2026 was conducted across six web-accessible scholarly search and publication sources using combinations of QSAR, molecular-property prediction, ADMET, explainability, attribution, counterfactual explanation, uncertainty quantification, calibration, conformal prediction, chemical-space generalization, and pharmaceutical decision terms. The review flow comprised 1,311 identified records, 342 retained after title screening, 87 retained after abstract assessment, and 29 finally included studies. Evidence was classified according to prediction context, explanation target, uncertainty formulation, validation strategy, chemical-space boundary, and relationship to pharmaceutical decision making. The literature showed that explanation quality cannot be inferred from predictive accuracy and that explanation outputs can vary with representation, model architecture, structural attribution unit, and benchmark design. Uncertainty estimates likewise differed across model classes, data distributions, activity-cliff regions, and calibration procedures. Counterfactual and contrastive approaches offered chemically explicit interrogation of prediction boundaries, but predicted structural alternatives did not constitute experimental causal evidence. Few studies directly connected joint explainability–uncertainty assessment to prospective pharmaceutical decisions. QSAR outputs become more defensible for pharmaceutical decision making when prediction, explanation, uncertainty, validation, and transferability are treated as distinct evidentiary layers. The review therefore develops a decision-oriented synthesis in which explanations require independent fidelity testing and uncertainty requires calibration within the chemical and decision domain in which it is used.
UR  - https://ijpras.com/article/explainable-and-uncertainty-aware-qsar-for-pharmaceutical-decision-making-iiy2bc7bnqqjyum
ER  -