Reproducibility package for controlled multi-regime evaluation of drug-target affinity models

Reproducibility release accompanying the manuscript:

"Model Rankings in Drug–Target Affinity Prediction Vary Across Evaluation Regimes: A Controlled Multi-Seed Study of Model-Comparison Reliability"

This release contains:

• frozen split indices and cryptographic hashes;
• source code for the three primary fusion models;
• controlled DeepDTA-style and GraphDTA-GCN-style baselines;
• formal experimental configurations;
• final run-level and aggregated results for all 126 prespecified evaluations;
• integrity-audit and held-out-test provenance materials;
• the exploratory VEGFR2 case-study analysis;
• the formal software-environment specification.

The original KIBA and Davis datasets are not redistributed in this repository.