{"success":true,"data":{"author":"Briney Lab, Scripps Research","bucketAccessType":"public","bucketName":"brineylab/oplm-antibody-sequences","bucketPrefix":"","category":"Biology","cloudProvider":"huggingface","createdAt":"2026-11-16T14:13:15.646Z","description":"Antibody language model training data from the Briney lab, built from AntiRef, its reference clustering of human antibody sequences. All 299,748,717 AntiRef99 sequences are grouped four ways, at 90, 92, 94 and 96 percent identity (antiref90 to antiref96 folders), with one row per cluster: the representative sequence and the member sequences inside it. Shards hold one million rows in a seeded global shuffle, so any subset is a uniform sample. Held-out files give evaluation and test sets of whole held-out clusters, plus a version 2.0.0 CDR3-identity holdout of 2,500 heavy chains each where no CDR3 in the holdout is within 85 percent identity of training data. Held-out rows carry locus and a per-residue CDR mask from IMGT numbering. Variable domains only, Parquet with zstd. CC BY 4.0.","displayTitle":"OPLM antibody sequences v2.0.0: AntiRef99 clustered training sets with CDR3 holdouts","huggingFaceUrl":"https://huggingface.co/datasets/brineylab/oplm-antibody-sequences","id":"c6fbb531-a49c-5d30-9fe4-233bf2184509","license":"CC-BY-4.0","licenseEvidenceUrl":"https://zenodo.org/records/7474336","objectCountEstimate":699,"probeStatus":"ok","referenceUrl":"https://zenodo.org/records/7474336","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"brineylab/oplm-antibody-sequences\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"brineylab/oplm-antibody-sequences\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=brineylab, license:cc-by-4.0, size_categories:100M<n<1B, format:parquet, downloads=106"],"format":"mixed","generatedAt":"2026-11-16T14:13:20.581Z","layout":{"example":"antiref90/train/shard-00000.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"antiref90/train/shard-00000.parquet","truncatedAt":512}],"sampleObjects":["antiref90/train/shard-00000.parquet","antiref90/train/shard-00001.parquet","antiref90/train/shard-00002.parquet","antiref90/train/shard-00003.parquet","antiref90/train/shard-00004.parquet"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\npretty_name: oplm-antibody-sequences\n---\n# oplm-antibody-sequences\n\nHub: `brineylab/oplm-antibody-sequences`, tag `v2.0.0`\n\nVersion 1.0.0 held out whole AntiRef90 clusters and nothing more, so 27% of eval_gte2's heavy chains had an exact copy of their CDR3 in train. Version 2.0.0 adds a holdout where \"unseen\" means unseen: no heavy-chain CDR3 left in training is 85% or more identical to a held-out heavy chain's CDR3. Its four train sets share that holdout and that screen, so OPLM decays that differ only in train set can be compared (briney/oplm#78). It is built from the published AntiRef (Briney 2023, *Bioinformatics Advances*, [10.1093/bioadv/vbad109](https://doi.org/10.1093/bioadv/vbad109)). A donor holdout waits on a rebuild from current OAS.\n\n## Files\n\n```\nantiref90/train/shard-00000.parquet …   one row per AntiRef90 cluster\nantiref92/train/ …  antiref94/train/ …  antiref96/train/ …\neval_cdrh3.parquet  test_cdrh3.parquet\neval_singletons.parquet  eval_gte2.parquet  test_singletons.parquet  test_gte2.parquet\n```\n\n- **`antiref{90,92,94,96}/train/`**, uniref70's columns:\n  - `sequence_id`, `sequence`: the cluster's representative, its AntiRef UUID and amino acids. **A row's `sequence` is AntiRef's representative unless the screen removed it**; then the cluster's first surviving member, in the seeded order, takes its place;\n  - `member_ids`, `member_sequences`: the surviving AntiRef99 representatives inside the cluster, the row's own entry first.\n\n  All four folders hold the same 299,748,717 AntiRef99 sequences, grouped differently, so the four train sets differ only in how training draws from them.\n- **Shards:** 1,000,000 rows each (the last fewer), row groups of 10,000, zstd. Rows are in a seeded global shuffle (`md5('1:<level id>')`), so any subset of shards is a uniform sample. Members follow the representative in `md5('1:<antiref99>')` order.\n- **`eval_cdrh3`, `test_cdrh3`:** 2,500 heavy chains each, the CDR3-identity holdout. Every CDR3 has at least 5 residues, and no two CDR3s across both files are 85% identical.\n- **`eval_singletons`, `eval_gte2`, `test_singletons`, `test_gte2`:** 1.0.0's files, byte-identical, so earlier #78 numbers stay comparable. 5,000 representatives of held-out AntiRef90 clusters each, heavy and light mixed.\n- **Held-out columns**, all six files:\n  - `sequence_id`, `sequence`;\n  - `locus`: `IGH`, `IGK`, or `IGL` (`IGH` only in the `_cdrh3` files);\n  - `cdr_mask_aa`: one digit per residue of `sequence`: `0` framework, `1` CDR1, `2` CDR2, `3` CDR3, from ANARCII 2.0.8's IMGT numbering (CDR1 27–38, CDR2 56–65, CDR3 105–117).\n\n  Every member of a held-out AntiRef90 cluster is out of every train set.\n\n`sequence` is the variable domain, FWR1 through FWR4.\n\n## Reading it\n\n```python\nfrom huggingface_hub import snapshot_download\npath = snapshot_download(\"brineylab/oplm-antibody-sequences\", repo_type=\"dataset\", revision=\"v2.0.0\")\n```\n\nIn OPLM, a `data.train` source is one of the four folders, **always with `sample_members: true`**: a row's representative may stand in for a removed one, so the representatives alone are not AntiRef's. The six held-out files are `type: sequence` evals; OPLM's `antibody_cdr3` eval reads them unchanged and scores only positions where `cdr_mask_aa` is `3`, per `locus`. Read heavy-only metrics on the `_cdrh3` files.\n\n## Recipe\n\n**Source:** AntiRef v2022.12.14 on Zenodo (CC BY 4.0), pulled 2026-11-09, the same two files as 1.0.0:\n- record 7474336, `antiref_cluster-manifest.csv.gz` (14,054,641,207 bytes, MD5 `18fcf6376702a2d23e2acdc32f1fd40a`): one row per AntiRef100 sequence, with its representative at 100, 99, 98, 96, 94, 92, and 90% identity;\n- record 7474658, `antiref100.fasta.gz` (18,496,427,276 bytes, MD5 `4a82d4be0c493377668fac9730aca866`): each sequence's UUID and amino acids.\n\nThe recipe refuses a download whose MD5 differs. AntiRef's clustering was MMseqs2 linclust, nested: each threshold clustered the previous one's representatives, with `-c 0.8 --cov-mode 0`.\n\n**Identity.** Two CDR3s `a` and `b` are 85% identical when `levenshtei","encoding":"utf8","key":"README.md","name":"README.md","size":10398,"truncated":true}],"sourceObjectKey":"antiref90/train/shard-00000.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"brineylab/oplm-antibody-sequences\")\nprint(ds)","starterSamplePath":"antiref90/train/shard-00000.parquet","totalObjects":699},"summary":"Antibody language model training data from the Briney lab, built from AntiRef, its reference clustering of human antibody sequences.","title":"OPLM antibody sequences v2.0.0: AntiRef99 clustered training sets with CDR3 holdouts","updatedAt":"2026-11-16T17:04:04.196Z","visibility":"public","websiteUrl":"https://huggingface.co/datasets/brineylab/oplm-antibody-sequences"}}