GVClass reference resources v1.5.0

Reference database archive for GVClass v1.5.0. What's new in v1.5.0 Contamination model retrained on simulated MAGs: the bundled ExtraTrees contamination regressor is trained on a 250-bin simulated-MAG dataset (200 realistically fragmented giant-virus MAGs with bacterial/eukaryotic/NCLDV-host-eukaryotic/viral-mixture contamination donors + 50 clean intact isolate genomes). Eliminates the novel-virus false-positive class observed in prior releases (mean predicted contamination on clean shredded NCLDV drops from ~29% to ~4%) while preserving calibration on intact curated inputs. Contaminated-bin MAE improves 3-5x across scenarios; Pearson r on contaminated bins reaches 0.97. External holdout on 50 unseen isolates: 0/50 exceed 10%. Reorganised layout: top-level grouped by concern (hmm/, markers/, completeness/, contamination/). Per-file naming dropped the novelty_strategy{2,3} prefix. The contamination model moved into resources/contamination/ so model rotation is a resources-tarball concern. Slimmed: dropped the vestigial database/dmnd/ directory (runtime uses pyswrd directly on .faa). Tarball size 3.2 GB -> 1.7 GB. Per-contig taxonomic-purity classifier: novel viruses with scattered cellular markers from HGT no longer downgrade to mixed_viral. Integrity: SHA-256 5357d96d99aa1eaf4b396ef701ed4c3b22d9015f79b7ae6c6be354c897704c80 Installation: pixi run setup-db (GVClass v1.5.0 points DEFAULT_DATABASE_SOURCE at this record). Concept DOI (all versions): 10.5281/zenodo.18662445