{"success":true,"data":{"author":"Adaptyv Bio","bucketAccessType":"public","bucketName":"adaptyvbio/proteinbase","bucketPrefix":"","category":"Biology","cloudProvider":"huggingface","createdAt":"2026-11-16T14:13:15.406Z","description":"Proteinbase is Adaptyv Bio's open corpus of designed proteins that were expressed and tested in its lab, published alongside computational predictions for the same sequences. Each record is one protein with a stable id, its amino-acid sequence, the design pipeline that produced it, the designer's handle when known, and a list of evaluations: experimental measurements such as binding to a target and expression, plus computed metrics, each with metric, method, value and unit. The snapshot tagged v2026-11-01 holds about 4,280 proteins as a single split named proteins, readable with load_dataset or as Parquet from the mount. It is the public record behind Adaptyv's binder design competitions and a benchmark for protein design models against wet-lab outcomes. Released under ODC-By 1.0.","displayTitle":"Proteinbase: experimentally tested protein designs from Adaptyv Bio (snapshot v2026-11-01)","huggingFaceUrl":"https://huggingface.co/datasets/adaptyvbio/proteinbase","id":"d64d59e5-3688-521e-9d70-6f5a09303a99","license":"ODC-By-1.0","licenseEvidenceUrl":"https://proteinbase.com/download","objectCountEstimate":3,"probeStatus":"ok","referenceUrl":"https://proteinbase.com","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"adaptyvbio/proteinbase\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"adaptyvbio/proteinbase\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=adaptyvbio, license:odc-by, size_categories:1K<n<10K, format:parquet, downloads=414"],"format":"mixed","generatedAt":"2026-11-16T14:13:17.515Z","layout":{"example":"data/proteins-00000-of-00001.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/proteins-00000-of-00001.parquet","truncatedAt":512}],"sampleObjects":["data/proteins-00000-of-00001.parquet"],"sidecars":[{"bytes":"---\nlicense: odc-by\ndataset_info:\n  features:\n  - name: id\n    dtype: string\n  - name: name\n    dtype: string\n  - name: sequence\n    dtype: string\n  - name: author\n    dtype: string\n  - name: designMethod\n    dtype: string\n  - name: evaluations\n    list:\n    - name: type\n      dtype: string\n    - name: metric\n      dtype: string\n    - name: valueType\n      dtype: string\n    - name: unit\n      dtype: string\n    - name: value\n      dtype: string\n    - name: target\n      dtype: string\n    - name: method\n      dtype: string\n    - name: measuredAt\n      dtype: string\n  splits:\n  - name: proteins\n    num_bytes: 34333362\n    num_examples: 4280\n  download_size: 4136369\n  dataset_size: 34333362\nconfigs:\n- config_name: default\n  data_files:\n  - split: proteins\n    path: data/proteins-*\nsize_categories:\n- 1K<n<10K\npretty_name: Proteinbase\ntags:\n- biology\n- proteins\n- protein-binding\n- ppi\n---\n# Proteinbase\n\nExperimentally validated protein-protein interaction data from\n[Adaptyv Bio](https://adaptyvbio.com), alongside computational predictions.\nPublished from <https://proteinbase.com/download>.\n\n```python\nfrom datasets import load_dataset\n\nds = load_dataset(\"adaptyvbio/proteinbase\")[\"proteins\"]\n\nfor ev in ds[0][\"evaluations\"]:\n    print(ev[\"metric\"], ev[\"method\"], ev[\"value\"], ev[\"unit\"])\n```\n\nThis is one unpartitioned corpus, so there is a single split named\n`proteins` rather than the usual `train`/`test` division.\n\nThis snapshot is tagged `v2026-11-01`. Pin it for reproducibility:\n\n```python\nds = load_dataset(\"adaptyvbio/proteinbase\", revision=\"v2026-11-01\")[\"proteins\"]\n```\n\n## Columns\n\n| Column | Description |\n|---|---|\n| `id` | Stable protein identifier, safe to treat as a primary key. |\n| `name` | Optional reference label; empty if none was supplied. |\n| `sequence` | Amino-acid sequence. |\n| `author` | Creator handle when available. |\n| `designMethod` | Slug of the design pipeline that generated the protein. |\n| `evaluations` | List of measurements for this protein (see below). |\n\n### `evaluations`\n\n| Field | Description |\n|---|---|\n| `type` | `computational`, `experimental`, or `curated`. |\n| `metric` | Metric slug, e.g. `kd`, `tm-score`, `expressed`. |\n| `target` | Binding target slug when applicable, otherwise empty. |\n| `valueType` | Parse hint: `numeric`, `boolean`, `label`, or `json`. |\n| `unit` | Unit symbol such as `nM`; empty when the metric is unitless. |\n| `value` | The measurement, encoded as a string (see note). |\n| `method` | Tool or protocol that produced it: `esmfold`, `boltz2`, `bli`, `spr`. |\n| `measuredAt` | ISO timestamp of the evaluation. |\n\n**`value` is always a string.** It is a union of number, boolean, label, and\nstructured JSON in the source data, and Arrow has no union type. Parse it\naccording to `valueType`: `numeric` with `float()`, `boolean` against\n`\"true\"`/`\"false\"`, `json` with `json.loads()`, `label` as-is.\n\n## When the same metric appears more than once\n\nComputational entries are deduplicated: where a tool has been run more than\nonce for the same protein and metric, only the most recent result is included.\n\nExperimental entries are **not** deduplicated. Several entries sharing a\n`metric` and `method` are genuine replicate measurements and should be\naggregated (e.g. averaged) rather than treated as duplicates.\n\nEntries sharing a `metric` but differing in `method` are different tools\nmeasuring the same quantity, and are never merged.\n\n## License\n\nOpen Data Commons Attribution License (ODC-By). Include this text in any work\nbuilding on Proteinbase:\n\n> \"This work used Proteinbase by Adaptyv Bio under ODC-BY license\"\n","encoding":"utf8","key":"README.md","name":"README.md","size":3606,"truncated":false}],"sourceObjectKey":"data/proteins-00000-of-00001.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"adaptyvbio/proteinbase\")\nprint(ds)","starterSamplePath":"data/proteins-00000-of-00001.parquet","totalObjects":3},"summary":"Proteinbase is Adaptyv Bio's open corpus of designed proteins that were expressed and tested in its lab, published alongside computational predictions for the same sequences.","title":"Proteinbase: experimentally tested protein designs from Adaptyv Bio (snapshot v2026-11-01)","updatedAt":"2026-11-16T17:01:45.537Z","visibility":"public","websiteUrl":"https://proteinbase.com"}}