CARDIOPROBE: Linear and non-linear probing of AlphaGenome embeddings

This dataset accompanies the manuscript: "Linear and non-linear probing of AlphaGenome embeddings recovers cardiac cell-type identity across the human autosome". Overview This deposit contains the processed data, trained models, and aggregate results required to reproduce every numerical claim, table, and figure in the CARDIOPROBE manuscript without re-running the heavy AlphaGenome encoder. The single source of truth for every quoted AUROC, gain, and statistic is located within the data.zip archive under stage2/method_aggregate/. File Contents data.zip: Contains interim curation decisions (mapping to 9 cell types), processed cell-type peak BEDs, per-bin GC fraction parquets across 22 autosomes, linear and non-linear panel JSONs, pooled-probe results, and trained linear-probe pickles (50/50-balanced). reports.zip: Contains data-curation and per-chromosome panel reports. MANIFEST.tsv: Contains per-file SHA-256 hashes and byte sizes for integrity verification. supplementary_data.pdf: Contains supplementary figures S1, S2 and tables S1, S2 and S3.