{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.02624", "kind": "paper", "url": "https://arxiv.org/abs/2606.02624"}, {"id": "gh:shiyegao/tadabench-1m", "kind": "repo", "url": "https://github.com/shiyegao/TadABench-1M"}, {"id": "hf:jingao/tadabench-1m", "kind": "dataset", "url": "https://huggingface.co/datasets/JinGao/TadABench-1M"}], "authors": ["Jin Gao", "Juntu Zhao", "Zirui Zeng", "Jiaqi Shen", "Junhao Shi", "Dukun Zhao", "Yuming Lu", "Dequan Wang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.02624", "source_url": "https://arxiv.org/abs/2606.02624"}, "categories": ["Biology & Drug Discovery", "Language & Knowledge", "q-bio.QM"], "description": {"en": "TadA-Bench is a fixed-data replay benchmark derived from 31 wet-lab rounds of TadA directed evolution. Models rank ~1M protein, DNA, or RNA sequence variants appearing only in later rounds, with scores as Spearman, Recall@10%, and nDCG@10%."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.02624", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.02624", "title": "TadA-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.02624", "modality": null, "name": "TadA-Bench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_tada-bench_769ea7c2", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "740b0600e60cc72cb34fb49a22af01b23eef4f884b99bdeab85666d76ee46d06", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.02624"}, "publisher": null, "released": "2026-05-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.02624", "source_url": "https://arxiv.org/abs/2606.02624"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-02624", "source": "claire_radar", "source_benchmark_id": "2606.02624", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["Biology & Drug Discovery"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/shiyegao/TadABench-1M", "githubScope": "benchmark_repo", "githubStars": 4, "hfDailySubmittedAt": null, "hfDataset": "JinGao/TadABench-1M", "hfDatasetDownloads": 143, "hfDatasetLikes": 0, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.02624", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}, "hfDatasetLikes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "a9967bd21cdc02003589cb6cb7db5ed19dcd7a95ea2a111f18834ae8f98ec307", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "TadA-Bench is a fixed-data replay benchmark derived from 31 wet-lab rounds of TadA directed evolution. Models rank ~1M protein, DNA, or RNA sequence variants appearing only in later rounds, with scores as Spearman, Recall@10%, and nDCG@10%.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce TadA-Bench, a million-variant wet-lab replay benchmark from 31 TadA directed-evolution rounds for future-round discovery toward agentic protein engineering."}, "familyId": "bmf_0784ab615c0b", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/shiyegao/TadABench-1M", "url": "https://github.com/shiyegao/TadABench-1M"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/shiyegao/TadABench-1M", "url": "https://huggingface.co/papers/2606.02624"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/shiyegao/TadABench-1M", "url": "https://arxiv.org/abs/2606.02624"}], "scope": "benchmark_repo", "url": "https://github.com/shiyegao/TadABench-1M", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_tada-bench_769ea7c2", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Pharma & Biotech"], "links": {"code": "https://github.com/shiyegao/TadABench-1M", "data": "https://huggingface.co/datasets/JinGao/TadABench-1M", "hfPaper": "https://huggingface.co/papers/2606.02624", "paper": "https://arxiv.org/abs/2606.02624", "pdf": "https://arxiv.org/pdf/2606.02624", "project": null, "report": "https://arxiv.org/abs/2606.02624"}, "metrics": [], "motivation": "AI for scientific discovery is entering an agentic era, where protein-engineering systems are expected to prioritize future wet-lab experiments rather than merely fit static measurements.", "name": "TadA-Bench", "oneLine": "TadA-Bench is a fixed-data replay benchmark derived from 31 wet-lab rounds of TadA directed evolution. Models rank ~1M protein, DNA, or RNA sequence variants appearing only in later rounds, with scores as Spearman, Recall@10%, and nDCG@10%.", "paperTitle": "TadA-Bench: A Million-Variant Benchmark for Future-Round Discovery Toward Agentic Protein Engineering", "primaryDomain": "Biology & Drug Discovery", "publication": {"evidence": "Accepted at the 43rd International Conference on Machine Learning (ICML 2026). Data: https://huggingface.co/datasets/JinGao/TadABench-1M . Code: https://github.com/shiyegao/TadABench-1M", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.02624", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "43rd International Conference on Machine Learning (ICML 2026)", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Shanghai Jiao Tong University", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/shiyegao/TadABench-1M"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-29", "source": {"authors": ["Jin Gao", "Juntu Zhao", "Zirui Zeng", "Jiaqi Shen", "Junhao Shi", "Dukun Zhao", "Yuming Lu", "Dequan Wang"], "categories": ["q-bio.QM", "cs.AI", "cs.LG"], "id": "2606.02624", "title": "TadA-Bench: A Million-Variant Benchmark for Future-Round Discovery Toward Agentic Protein Engineering", "type": "arxiv", "url": "https://arxiv.org/abs/2606.02624"}, "sourceUpdatedAt": "2026-05-29T12:12:08Z", "topics": ["q-bio.QM"], "venueAttempts": [{"decisionRaw": "Accepted at the 43rd International Conference on Machine Learning (ICML 2026). Data: https://huggingface.co/datasets/JinGao/TadABench-1M . Code: https://github.com/shiyegao/TadABench-1M", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted at the 43rd International Conference on Machine Learning (ICML 2026). Data: https://huggingface.co/datasets/JinGao/TadABench-1M . Code: https://github.com/shiyegao/TadABench-1M", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.02624"}], "reviewStatus": "accepted", "venueName": "43rd International Conference on Machine Learning (ICML 2026)"}], "whyItMatters": "Standard random-split evaluation overestimates performance for iterative candidate prioritization. TadA-Bench provides a chronological replay protocol to measure whether models can transfer from earlier experimental rounds to future ones, a core requirement for agentic protein engineering."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
