protein-ligand-affinity-web-assets

Protein–ligand affinity website resources

Large, immutable input files for the HKMU AI4S Lab project website, project UGC/FDS16/E16/23. The website source and validated inference manifests are maintained in HKMU-AI4S-Lab/protein-ligand-affinity.

The content-addressed weight file is an unchanged FP32 initializer exported from the full MAO-B DL-FSG/Fusion screening checkpoint. It is used with the smaller ONNX files and weight arrays distributed with the website. See Deep learning on the fusion of chemical sequences and molecular grids. The research owner authorized publication of the website and its selected inference assets on 25 September 2026. The upstream MCANet BSD-3-Clause notice is retained in FUSION-UPSTREAM-LICENSE.txt; it is not a blanket licence for unrelated research materials.

The two grid files are prepared inputs for the external GNINA crossdock_default2018 model, using the 1HSG and 1HVR crystallographic complexes and libmolgrid 0.5.5. GNINA and gnina-torch are external software, credited in the website and its notices. See GNINA 1.0 and gnina-torch. These grids are input tensors, not experimental measurements or new trained models.

The website pins a complete repository commit, verifies every file's byte count and SHA-256 before inference, and caches completed downloads in the visitor's browser. Files are downloaded only when the visitor explicitly starts the corresponding analysis.