TY  - JOUR
T1  - Rebuilding Quantitative Structure–Activity Relationship Modeling around Chemical Neighborhoods, Domain Boundaries, and Decision-Relevant Reliability
A1  - Lucas Pereira
A1  - Carolina Mendez
A1  - Felipe Rios
A1  - David Taylor
JF  - Pharmacophore
JO  - Pharmacophore
SN  - 2229-5402
Y1  - 2024
VL  - 15
IS  - 6
DO  - 10.51847/xPQxd9V8T7
SP  - 57
EP  - 65
N2  - Quantitative structure–activity relationship modeling remains central to computational pharmaceutical science, yet its reliability is commonly summarized through aggregate performance measures that conceal where, why, and for which decisions predictions are scientifically supportable. A model may perform favorably across a retrospective test set while remaining unreliable for a sparsely represented chemical series, a locally discontinuous activity landscape, an unfamiliar molecular modality, or a decision carrying asymmetric experimental consequences. This article develops a methodological reconstruction of QSAR evaluation around chemical neighborhoods, explicit domain boundaries, calibrated uncertainty, and decision-relevant reliability. The approach synthesizes evidence concerning benchmark design, molecular representation, applicability-domain assessment, activity cliffs, local predictability, uncertainty estimation, external validation, and reproducible reporting. Its central conceptual contribution is a proposed Neighborhood–Domain–Decision Reliability framework in which reliability is assigned not to a model globally, but conditionally to a prediction–decision pair. The framework distinguishes representation-specific neighborhood support, local activity-landscape behavior, multidimensional applicability, uncertainty calibration, and intended decision consequences. It further proposes reliability tiers separating supported interpolation, boundary-adjacent or structured extrapolation, and out-of-domain use requiring abstention or hypothesis-generating interpretation. The reconstruction does not claim validated universal thresholds, prospective pharmaceutical utility, mechanistic explanation, regulatory acceptance, or deployment readiness. Instead, it offers a testable structure for designing evaluations, reporting limitations, and directing future validation. Rebuilding QSAR around these elements may improve the scientific interpretability of predictions, reduce false precision, and clarify when computational outputs can cautiously support compound prioritization, additional evidence generation, or explicit non-use.
UR  - https://pharmacophorejournal.com/article/rebuilding-quantitative-structureactivity-relationship-modeling-around-chemical-neighborhoods-doma-2a9fclezawzf2q7
ER  -