Multi-target VS benchmark — per-target data deposit (v1.0)

Companion Zenodo Dataset deposit for the manuscript "A controlled multi-target benchmark of skill-file-guided agentic virtual screening across eleven drug targets" (Gani, 2026, GigaScience Technical Note submission), and for the lean code repository at https://github.com/oslo-medchem/vs-multitarget-benchmark (concept DOI 10.5281/zenodo.20036546).

  This deposit contains the per-target benchmark data that was split out from the code repository to keep the GitHub auto-archive lean: ChEMBL-derived 100-active sets, property-matched decoys, receptor and ligand PDBQT files, Uni-Dock score tables, and per-target evaluation outputs (ROC, BEDROC, EF, AUPR, LogAUC) for all 11 protein targets used in the study (FPR2, EGFR, CDK2, COX-2, ESR1, DPP-4, AChE, BACE-1, HSP90-alpha, p38 MAPK, Thrombin).

To re-execute the analysis end to end: clone the GitHub repo, download this zip, unzip it into the repository root so the targets/ directory is restored alongside common/, prompts/, 06_meta_analysis/, and 07_manuscript/, then follow the run instructions in the repo's CLAUDE.md.