AlphaGenome embedding probing: data products for cardiac chromatin accessibility analysis

Derived data products supporting "Linear probing of AlphaGenome embeddings recovers tissue-specific regulatory state: a stage-matched, baseline-controlled evaluation" (Astudillo, BIOADV-2026-321, Bioinformatics Advances). Contents (22 autosomes unless noted): genome-binned ENCODE DNase labels (including the stage-matched 2x2, held-out-biosample, and designed-panel variants); AlphaGenome 305-track DNase head predictions (chr1/11/22); per-tissue, held-out, adult-negative, and panel peak-to-bin overlaps with provenance; fold-specific embeddings for the held-out-region analysis; H3K27ac modality comparison (chr1/11/22); result JSONs for every analysis; final manuscript figures and table sources; track metadata (life stage, ontology); and MANIFEST.tsv with path, size, and SHA-256 for every file. Not included: the raw AlphaGenome embeddings (~600 GB), which are deterministic given the model and genomic bin coordinates and reproducible from the deposited code plus AlphaGenome access. Code: 10.5281/zenodo.21948978 (concept DOI) / github.com/patricio-astudillo/alphagenome-embedding-probing. No new biological data were generated; all source peaks are from ENCODE.