Reproducibility Package for "Scoring-Function Terms Can Reverse Docking-Derived Descriptor Direction: An Analytical Criterion and Prospective Vina Confirmation from a Paired Caffeine Library"

This repository contains the reproducibility materials supporting the manuscript “Scoring-Function Terms Can Reverse Docking-Derived Descriptor Direction: An Analytical Criterion and Prospective Vina Confirmation from a Paired Caffeine Library.” The study investigates whether directional molecular-descriptor interpretations derived from docking-score surrogate models can be altered by ligand-dependent terms in the scoring function. The deposited materials support analyses of a curated paired library of 809 C8-substituted caffeine derivatives evaluated against monoamine oxidase B (MAO-B) and the adenosine A2A receptor. The archive contains the curated target-specific and paired molecular datasets, the prospective AutoDock Vina 1.2.7 confirmatory output for all 1,618 ligand–target calculations, the Vina-ready TORSDOF audit for all 809 ligands, machine-readable recalculation of the engine-level direction test reported in Table 8, ChEMBL-derived experimental reference datasets and reconstruction code, and SHA-256 integrity information. The prospective confirmatory experiment completed 1,618/1,618 ligand–target calculations successfully. Direct subtraction of the engine-reported torsional contribution reverses the flexibility–score association at both MAO-B and A2A, whether flexibility is represented by the library rotatable-bond descriptor or by Vina-ready TORSDOF. The deposited raw confirmatory output permits independent recalculation of these results. The ChEMBL-derived datasets are included as chemically heterogeneous experimental reference material and are not presented as same-chemotype biological validation. ChEMBL-derived content remains subject to the applicable ChEMBL licensing terms. Underlying Reaxys records are not redistributed; compounds in the deposited curated datasets are represented by canonical molecular structures generated in-house under the institutional Reaxys licence. The enumerated prospective analogues and measured-activity model outputs are hypothesis-generating computational results and should not be interpreted as experimentally validated potency predictions. Software environment used for the prospective confirmatory docking: AutoDock Vina 1.2.7, Meeko 0.7.1, RDKit 2025.09.5 and Gemmi 0.7.5. Corresponding authors: Gülbin Kurtay ([e-mail]) and Fatma Sevin ([e-mail]).