dtSFM v3 — candidate pools and AlphaFold 3 verification

Supplementary data for the drug–target specificity foundation model (dtSFM) : A Drug–Target Specificity Foundation Model for Off-target Prediction, Repurposing, and Generative Design. This deposit contains the large data artifacts that back the paper's claims but are too large for the GitHub code repository (github.com/Reddy-BIIE-ETHZ/dtSFM): the full library-repurposing rankings, the generative-design candidate cohort with per-design SMILES and AlphaFold-3 structural-confidence profiles, the AlphaFold-3 cofold structures (CIF + summary confidences) for representative complexes, the affinity-calibration cohorts, and the atpE pilot deliverable. Model weights are distributed separately on Hugging Face (SFM-BIIE-ETHZ/dtSFM-v3). All numerical claims are traceable via docs/reproducing_paper_claims.md in the code repository; the closed orthogonal (Codex) audit is in audit/. The specific molecules disclosed here are dedicated to the public.