Release Notes

Version 1.1.0

Added mouse reference support across config generation, preprocessing, prediction, and ATAC-to-RNA link inference.
Added --species {human,mouse} to species-sensitive commands.
Added species-aware config generation. Human configs use total_gene: 38244; mouse configs use total_gene: 23234.
Genome annotation .gtf.gz files are no longer bundled in the PyPI package. generate_default_config downloads the selected species Gencode annotation into cisformer_config/resource/.

ATAC-to-RNA preprocessing now performs automatic log1p detection. Cisformer checks the loaded RNA matrix and prints whether scanpy.pp.log1p is applied.
Improved preprocessing output buffering.
Improved atac2rna_link performance by reducing repeated Python-level peak lookup and sparse matrix writes.
Documentation now clarifies that atac2rna_link writes rank-normalized link scores rather than raw correlation coefficients.

Regenerate configs with cisformer generate_default_config --species human or --species mouse before training new v1.1.0 models.
Use the same species value for preprocessing, prediction, and link generation.
Keep cisformer_config/resource/ with your project. It contains downloaded genome annotations and generated link-neighborhood caches.
Existing human models trained with the old 38,244-gene vocabulary should use human configs.