CyanoDiff: data, model checkpoints, and code for class-conditional cyanobacterial promoter generation via masked diffusion language modeling

Reproduction bundle for the paper "CyanoDiff: Class-Conditional Cyanobacterial Promoter Generation via Masked Diffusion Language Modeling". CyanoDiff is a masked diffusion language model (MDLM) for designing cyanobacterial promoters with controllable expression strength. This deposit contains: data.tar.gz: the 486-genome pretraining corpus (FASTA, GFF, 16S, metadata) and per-species processed datasets (train/val/test, GMM expression bins) for Nostoc sp. PCC 7120, Synechocystis sp. PCC 6803, Prochlorococcus MED4, and MIT9313, plus raw source data and reference genomes. checkpoints.tar.gz: the core trained weights — a genome-pretrained model and four species fine-tuned models. code.tar.gz: the core model, training, and sampling code (identical to the GitHub repository), provided as a self-contained, citable snapshot. Code is maintained at https://github.com/Passion4ever/CyanoDiff