{"success":true,"data":{"author":"Bernett, Blumenthal and List (TUM daisybio); packaged by Synthyra","bucketAccessType":"public","bucketName":"Synthyra/bernett_gold_ppi","bucketPrefix":"","category":"Proteomics","cloudProvider":"huggingface","createdAt":"2026-11-14T14:15:14.537Z","description":"Parquet packaging of the gold standard dataset for sequence-based protein-protein interaction (PPI) prediction from Bernett, Blumenthal and List, Cracking the black box of deep sequence-based protein-protein interaction prediction (Briefings in Bioinformatics 2024). The benchmark was built to remove the data leakage that inflates published PPI accuracies: the human proteome was partitioned with KaHIP so that sequence similarity between partitions is minimised, proteins in training never appear in validation or test, and CD-HIT removed redundancy above 40 percent pairwise identity inside each set. Positives come from the HIPPIE interaction database and sequences from Swiss-Prot; the data are balanced binary labels. Splits: 163,192 training pairs (Intra-1), 59,260 validation pairs (Intra-0) and 52,048 test pairs (Intra-2). Each row has the two UniProt accessions (A, B), a label and both amino acid sequences (SeqA, SeqB). Measured 2026-11-14: three parquet files under data/ totalling about 0.29 GB. Load with datasets.load_dataset('Synthyra/bernett_gold_ppi') or pandas. The authors' follow-up study reports that deep models on this benchmark plateau near 0.65 accuracy, which makes it a hard test for protein language model embeddings and PPI classifiers. The original figshare record is licensed CC BY 4.0; cite Bernett et al. 2024.","displayTitle":"Leakage-Free Gold Standard PPI Dataset (Bernett et al.)","doi":"10.6084/m9.figshare.21591618.v4","id":"3bdb3b5a-48ef-5138-ac52-91168829d59e","license":"CC-BY-4.0","licenseEvidenceUrl":"https://api.figshare.com/v2/articles/21591618","objectCountEstimate":5,"probeStatus":"ok","referenceUrl":"https://doi.org/10.1093/bib/bbae076","region":"us","schemaCard":{"accessHints":[{"language":"python-datasets","snippet":"from datasets import load_dataset\nds = load_dataset(\"Synthyra/bernett_gold_ppi\")"},{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"Synthyra/bernett_gold_ppi\", repo_type=\"dataset\")"}],"caveats":["HuggingFace metadata: author=Synthyra, license:cc-by-4.0, size_categories:100K<n<1M, format:parquet, downloads=118"],"format":"mixed","generatedAt":"2026-11-14T14:15:19.765Z","layout":{"example":"data/test-00000-of-00001.parquet","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleHeaders":[{"bytes":"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","encoding":"hex","key":"data/test-00000-of-00001.parquet","truncatedAt":512}],"sampleObjects":["data/test-00000-of-00001.parquet","data/train-00000-of-00001.parquet","data/valid-00000-of-00001.parquet"],"sidecars":[{"bytes":"---\nlicense: cc-by-4.0\ntask_categories:\n- text-classification\ntags:\n- protein\n- biology\n- protein-protein-interaction\n- synthyra\ndataset_info:\n  features:\n  - name: A\n    dtype: string\n  - name: B\n    dtype: string\n  - name: labels\n    dtype: int8\n  - name: SeqA\n    dtype: string\n  - name: SeqB\n    dtype: string\n  splits:\n  - name: train\n    num_bytes: 271796192\n    num_examples: 163192\n  - name: valid\n    num_bytes: 73318294\n    num_examples: 59260\n  - name: test\n    num_bytes: 57573817\n    num_examples: 52048\n  download_size: 291089780\n  dataset_size: 402688303\nconfigs:\n- config_name: default\n  data_files:\n  - split: train\n    path: data/train-*\n  - split: valid\n    path: data/valid-*\n  - split: test\n    path: data/test-*\n---\n\n# Leakage-free \"gold\" standard PPI dataset\n\nFrom [Bernett, et al](https://figshare.com/articles/dataset/PPI_prediction_from_sequence_gold_standard_dataset/21591618), found in \n\nCracking the black box of deep sequence-based protein–protein interaction prediction\n* [paper](https://academic.oup.com/bib/article/25/2/bbae076/7621029)\n* [code](https://github.com/daisybio/data-leakage-ppi-prediction)\n\nand \n\nDeep learning models for unbiased sequence-based PPI prediction plateau at an accuracy of 0.65\n* [paper](https://www.biorxiv.org/content/10.1101/2025.01.23.634454v1)\n* [code](https://github.com/daisybio/PPI_prediction_study)\n\n## Description\nThis is a balanced binary protein-protein interaction dataset with positives from [HIPPIE](https://cbdm-01.zdv.uni-mainz.de/~mschaefer/hippie/) and partitioned with [KaHIP](https://kahip.github.io/). There are no sequence overlaps in splits, furthermore, they are split based on a maximum CD-HIT of 40% pairwise sequence similarity. Node degree bias was also reduced.\n\nNote:\n\n[Q96PU5](https://www.uniprot.org/uniprotkb/Q96PU5) is not located in the provided SwissProt fasta file but is used in the train split several times. We added this before data processing so no rows were dropped.\n\n## Example use\n```python\ndef get_ppi_data():\n    data = load_dataset(\"Synthyra/bernett_gold_ppi\").shuffle(seed=42)\n    data = data.remove_columns(['A', 'B'])\n    return data['train'], data['valid'], data['test']\n```\n\n## Please cite\nPlease cite their figshare dataset and papers if you use this dataset. \n","encoding":"utf8","key":"README.md","name":"README.md","size":2278,"truncated":false}],"sourceObjectKey":"data/test-00000-of-00001.parquet","starterCell":"from datasets import load_dataset\nds = load_dataset(\"Synthyra/bernett_gold_ppi\")\nprint(ds)","starterSamplePath":"data/test-00000-of-00001.parquet","totalObjects":5},"summary":"Parquet packaging of the gold standard dataset for sequence-based protein-protein interaction (PPI) prediction from Bernett, Blumenthal and List, Cracking the black box of deep…","title":"Leakage-Free Gold Standard PPI Dataset (Bernett et al.)","updatedAt":"2026-11-14T17:47:13.718Z","visibility":"public","websiteUrl":"https://figshare.com/articles/dataset/PPI_prediction_from_sequence_gold_standard_dataset/21591618"}}