vHold: Supplementary Models and Databases

Supplementary data files for vHold (https://github.com/shandley/vhold), a viral protein annotation tool using structural homology (ProstT5 + Foldseek). This record contains: - vhold_embeddings.npz (784 MB): Pre-computed ProstT5 encoder embeddings for 436,237 viral protein structures from BFVD and Viro3D. Used for fast embedding-based triage. - bfvd_metadata_enriched.tsv (79 MB): Enriched BFVD metadata with UniProt annotations (protein names, GO terms, Pfam domains, keywords, taxonomic lineage) for 345,730 proteins. - vhold_classifier.pt (2.5 MB): Trained MLP functional classifier for viral protein category prediction (macro F1 0.692). Install via: vhold install --embeddings --classifier