{"success":true,"data":{"author":"Dmytro Rizdvanetskyi, Nathan Roos, Pavlo Lutsik (KU Leuven)","bucketAccessType":"public","bucketName":"CompEpigen/syto.1.0","bucketPrefix":"","category":"Genomics & Bioinformatics","cloudProvider":"huggingface","createdAt":"2026-11-21T20:45:58.947Z","description":"Data underlying the paper Data-driven soft labeling scales DNA read classification to whole-body cell-type deconvolution. SYTO is a framework that classifies individual bisulfite sequencing reads by their DNA methylation pattern to estimate the cell-type composition of mixed DNA samples (cell-type deconvolution) across cell types of the whole human body; source reads include recovered reads from the Loyfer et al. methylation atlas. The deposit contains everything behind the paper's figures and tables in five tiers: tier0-results (final result tables, 0.5 GB, enough to inspect the findings), tier1-models (trained models for each benchmark run, including RRBS out-of-distribution runs), tier2-pseudobulks (generated in-silico pseudobulk mixtures), tier3-training-data and tier4-source-data (staged and recovered reads plus hg19 and hg38 references). 190 GB in 84 zip archives with MANIFEST.csv and MANIFEST_CONTENTS.csv describing the 82,010 unpacked files; fetch selected tiers with hf download --include. Mirror of the KU Leuven Research Data Repository record, DOI 10.48804/JM4BOS, CC BY 4.0. Cite Rizdvanetskyi, Roos and Lutsik.","displayTitle":"SYTO v1.0: Read-Level DNA Methylation Classifiers and Pseudobulk Mixtures for Whole-Body Cell-Type Deconvolution (KU Leuven)","doi":"10.48804/JM4BOS","huggingFaceUrl":"https://huggingface.co/datasets/CompEpigen/syto.1.0","id":"5661f37b-348b-5f27-b1f7-d1bdb0e5668c","license":"CC-BY-4.0","licenseEvidenceUrl":"https://rdr.kuleuven.be/api/datasets/:persistentId/?persistentId=doi:10.48804/JM4BOS","objectCountEstimate":91,"probeStatus":"ok","referenceUrl":"https://huggingface.co/datasets/CompEpigen/syto.1.0","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"CompEpigen/syto.1.0\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"CompEpigen/syto.1.0\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=CompEpigen, license:cc-by-4.0, downloads=191"],"format":"mixed","generatedAt":"2026-11-21T20:48:49.308Z","layout":{"example":"MANIFEST.csv","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"item,tier,description,bytes,files,sha256\r\ntier0-results/SYTO_tier0_results_baselines_v1.zip,tier0,Final result tables: baselines,381705958,37,1d6f5bc7bebeda1af11016f3c579627afde81b7efdcc6c61bcd094d963f6aedb\r\ntier0-results/SYTO_tier0_results_cfsortooddeconvolutionresults_v1.zip,tier0,Final result tables: cfsort_ood_deconvolution_results,72581474,22,8c260e3e79419b20ce4e64a9ea434145f34e8a38f4e31f67417f9d0eae3aa9e0\r\ntier0-results/SYTO_tier0_results_mappings_v1.zip,tier0,Final result tables: mappings,10872,4,89f","encoding":"utf8","key":"MANIFEST.csv","truncatedAt":512}],"sampleObjects":["FILE_NAMING_CONVENTION.txt","MANIFEST.csv","MANIFEST_CONTENTS.csv","runs.csv","tier0-results/SYTO_tier0_results_baselines_v1.zip"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: \"SYTO v1.0 — data-driven soft labeling for whole-body cell-type deconvolution\"\nviewer: false\ntags:\n  - biology\n  - genomics\n  - epigenomics\n  - dna-methylation\n  - cell-type-deconvolution\n  - read-classification\n---\n\n# SYTO v1.0 publication data\n\nData underlying *\"Data-driven soft labeling scales DNA read classification to\nwhole-body cell-type deconvolution\"*. SYTO is a framework for read-level DNA\nmethylation classification and whole-body cell-type deconvolution. This deposit\ncontains the trained models, generated pseudobulk mixtures, training data,\nsource reads and final result tables behind every figure and table in the paper.\n\n> **This repository is a mirror.**\n> The citable version of the record is deposited in the Research Data Repository of KU Leuven, doi:10.48804/JM4BOS.\n> This mirror exists to make the data easier\n> to fetch programmatically; contents are identical to the planed citable deposit.\n\n- **Size:** 190.4 GB across 89 files (84 `.zip` archives + 4 acompanying metadata files + 1 csv file with Syto variants and baselines ranking)\n- **Unpacked:** 82,010 files\n- **Largest file:** 5.66 GB\n\n## Contents\n\nThe deposit is organised in five tiers of decreasing necessity, so you can take\nonly the depth you need.\n\n| Tier | Contents | Size |\n|---|---|---|\n| `tier0-results/` | Final result tables underlying important published figures and tables | 0.5 GB |\n| `tier1-models/` | Trained classifiers, deconvolvers and calibrators | 24.0 GB |\n| `tier2-pseudobulks/` | Generated pseudobulk mixtures (columnar Parquet) | 141.5 GB |\n| `tier3-training-data/` | Marker atlases, target proportions, training datasets | 12.6 GB |\n| `tier4-source-data/` | Staged and recovered reads, hg19/hg38 reference genomes | 11.8 GB |\n\n**`tier0-results/` alone (0.5 GB) is enough to inspect the final results of the paper.**\nTiers 1–4 exist so the analysis can be independently re-run and verified.\n\nMetadata files are readable without downloading anything large:\n\n| File | Purpose |\n|---|---|\n| `MANIFEST.csv` | One row per archive: description, size, file count, SHA-256 |\n| `MANIFEST_CONTENTS.csv` | One row per file *inside* the archives — inspect an archive's contents without downloading it |\n| `runs.csv` | One row per experiment run, linking tier1 ↔ tier2 ↔ tier3 paths |\n| `FILE_NAMING_CONVENTION.txt` | Full naming scheme and every abbreviation used |\n\n## Downloading\n\nArchives are stored as-is; there is no dataset viewer. Fetch selectively.\n\nInspect what exists first — both manifests are small:\n\n```python\nimport pandas as pd\npd.read_csv(\"hf://datasets/CompEpigen/syto.1.0/MANIFEST.csv\")\npd.read_csv(\"hf://datasets/CompEpigen/syto.1.0/MANIFEST_CONTENTS.csv\")\n```\n\nJust the results (0.5 GB):\n\n```bash\nhf download CompEpigen/syto.1.0 --repo-type=dataset \\\n  --include \"tier0-results/*\" \"*.csv\" \"*.txt\" --local-dir ./syto-data\n```\n\nOne specific run:\n\n```bash\nhf download CompEpigen/syto.1.0 --repo-type=dataset \\\n  --include \"tier1-models/ood-rrbs/SYTO_tier1_models_oodrrbs_dismir_softlabelpooled_v1.zip\" \\\n  --local-dir ./syto-data\n```\n\nEverything (190 GB):\n\n```bash\nhf download CompEpigen/syto.1.0 --repo-type=dataset --local-dir ./syto-data\n```\n\n## Unpacking\n\n**Always unpack from the deposit root.** Archive members are stored with paths\nrelative to the root, so unpacking an archive from inside its own folder nests\nthe tree a second time (`tier1-models/ood-rrbs/tier1-models/...`) \n\n```bash\ncd ./syto\nfind . -name 'SYTO_*.zip' -exec unzip -o -q {} ';'\n```\n\nThe reference-genome archives contain relative symlinks\n(`genome.fa.gz -> hg38.fa.gz`) matching the `wgbs_tools` layout. Unpack them\nwith a tool that preserves symlinks — `unzip` on Linux and macOS does.\n\n## Verifying\n\n`MANIFEST.csv` carries a SHA-256 for every archive:\n\n```bash\npython - <<'PY'\nimport csv, hashlib, pathlib\nfor r in csv.DictReader(open(\"MANIFEST.csv\")):\n    p = pathlib.Path(r[\"item\"])\n    if not p.exists():\n        continue\n    h = hashlib.sha256()\n    with p.open(\"rb\") as f:\n        for chunk in iter(lambda: f.read(1 << 20), b\"\"):\n            h.upd","encoding":"utf8","key":"README.md","name":"README.md","size":6293,"truncated":true}],"sourceObjectKey":"MANIFEST.csv","starterCell":"import pandas as pd\ndf = pd.read_csv(\"hf://datasets/CompEpigen/syto.1.0/MANIFEST.csv\")\ndf.head()","starterSamplePath":"MANIFEST.csv","totalObjects":91},"summary":"Data underlying the paper Data-driven soft labeling scales DNA read classification to whole-body cell-type deconvolution.","title":"SYTO v1.0: Read-Level DNA Methylation Classifiers and Pseudobulk Mixtures for Whole-Body Cell-Type Deconvolution (KU Leuven)","updatedAt":"2026-11-22T01:03:14.602Z","visibility":"public","websiteUrl":"https://doi.org/10.48804/JM4BOS"}}