PhaGen: code, evaluation data and results

Supporting materials for “Masked diffusion for bacteriophage genome analysis and fast genome-scale generation”. PhaGen is a masked diffusion model for generating and analyzing bacteriophage DNA.

The archive includes PhaGen model weights and tokenizer files, the model implementation, examples, VeOmni-based training code, evaluation inputs and processed benchmarks, formal results, analysis and plotting scripts, and figures for genome generation and seven sequence-understanding tasks: lambda/P1 gene essentiality, T7 SNP mutability, phage promoter recognition, virion-protein subtype prediction, tail-protein CDS host prediction, infA fitness and TEM-1 beta-lactamase fitness.

Extract PhaGen-complete.zip and follow its README and task documentation. PhaGen model weights and tokenizer files are included in PhaGen/models/PhaGen/; baseline model weights and full external reference databases are obtained separately. Data sources and applicable upstream terms are recorded in PhaGen/evaluation/DATA_SOURCES.md; software attribution is in PhaGen/NOTICE.md. No new license is assigned here to project-authored predictions, statistical summaries or figures.