{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.04525", "kind": "paper", "url": "https://arxiv.org/abs/2606.04525"}, {"id": "gh:darlednik/geneb", "kind": "repo", "url": "https://github.com/darlednik/GENEB"}], "authors": ["Daria Ledneva", "Mikhail Nuridinov", "Denis Kuznetsov"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.04525", "source_url": "https://arxiv.org/abs/2606.04525"}, "categories": ["Biology & Drug Discovery", "Language & Knowledge", "cs.CL"], "description": {"en": "GENEB evaluates frozen representations from 40 genomic foundation models across 100 DNA classification tasks in 13 functional categories, using a unified linear probing protocol with full-data, 10-shot, and 1-shot regimes. Primary metric is Matthews correlation coefficient, with rankings at overall, category, and task levels."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.04525", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.04525", "title": "GENEB"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.04525", "modality": null, "name": "GENEB", "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_geneb_31b4a4f4", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "3eec38309691709ce5a703c1183237c92e03b45c669c8510b87c7320c3eed948", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.04525"}, "publisher": null, "released": "2026-06-03", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.04525", "source_url": "https://arxiv.org/abs/2606.04525"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-04525", "source": "claire_radar", "source_benchmark_id": "2606.04525", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Mixed", "applicationDomains": ["Biology & Drug Discovery"], "area": "Language & Knowledge", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/darlednik/GENEB", "githubScope": "benchmark_repo", "githubStars": 46, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 50, "hfPaperUrl": "https://huggingface.co/papers/2606.04525", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-25", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Aggregate Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "0d0ed786690b2a64905443e202f1633195faa91195eaebe0bcbc6a44b65c2915", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "GENEB evaluates frozen representations from 40 genomic foundation models across 100 DNA classification tasks in 13 functional categories, using a unified linear probing protocol with full-data, 10-shot, and 1-shot regimes. Primary metric is Matthews correlation coefficient, with rankings at overall, category, and task levels.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "We introduce GENEB, a large-scale diagnostic benchmark that evaluates frozen representations from 40 genomic foundation models across 100 tasks spanning 13 functional categories under a unified probing-based protocol, including few-shot regimes."}, "familyId": "bmf_85ca5f7ab2ea", "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/darlednik/GENEB", "url": "https://github.com/darlednik/GENEB"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/darlednik/GENEB", "url": "https://huggingface.co/papers/2606.04525"}], "scope": "benchmark_repo", "url": "https://github.com/darlednik/GENEB", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_geneb_31b4a4f4", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": ["Pharma & Biotech"], "links": {"code": "https://github.com/darlednik/GENEB", "data": null, "hfPaper": "https://huggingface.co/papers/2606.04525", "paper": "https://arxiv.org/abs/2606.04525", "pdf": "https://arxiv.org/pdf/2606.04525", "project": null, "report": "https://arxiv.org/abs/2606.04525"}, "metrics": [], "motivation": "Progress in genomic foundation models is difficult to assess due to fragmented benchmarks, incompatible evaluation protocols, and task-specific reporting.", "name": "GENEB", "oneLine": "GENEB evaluates frozen representations from 40 genomic foundation models across 100 DNA classification tasks in 13 functional categories, using a unified linear probing protocol with full-data, 10-shot, and 1-shot regimes. Primary metric is Matthews correlation coefficient, with rankings at overall, category, and task levels.", "paperTitle": "GENEB: Why Genomic Models Are Hard to Compare", "primaryDomain": "Biology & Drug Discovery", "publication": {"evidence": "Accepted to ICML 2026", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.04525", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "ICML 2026", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "GENEB team", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/darlednik/GENEB"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-03", "source": {"authors": ["Daria Ledneva", "Mikhail Nuridinov", "Denis Kuznetsov"], "categories": ["cs.CL", "cs.LG", "q-bio.GN"], "id": "2606.04525", "title": "GENEB: Why Genomic Models Are Hard to Compare", "type": "arxiv", "url": "https://arxiv.org/abs/2606.04525"}, "sourceUpdatedAt": "2026-08-05T10:47:21Z", "topics": ["cs.CL"], "venueAttempts": [{"decisionRaw": "Accepted to ICML 2026", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted to ICML 2026", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.04525"}], "reviewStatus": "accepted", "venueName": "ICML 2026"}], "whyItMatters": "Genomic model comparisons are fragmented across incompatible protocols, making claims of superiority unreliable. GENEB provides a controlled, multi-task reference for category-aware model selection, revealing that aggregate leaderboards are unstable and scale gains are inconsistent."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
