{"id":"W4408083017","doi":"10.1101/2025.02.25.640181","title":"geneRNIB: a living benchmark for gene regulatory network inference","year":2025,"lang":"en","type":"preprint","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Gene Regulatory Network Analysis","field":"Biochemistry, Genetics and Molecular Biology","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada); Mila - Quebec Artificial Intelligence Institute; Université de Montréal","funders":"","keywords":"Inference; Benchmark (surveying); Gene regulatory network; Computational biology; Gene; Computer science; Biology; Genetics; Artificial intelligence; Gene expression; Geography; Cartography","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0011358,0.0009393925,0.0009138078,0.0002402946,0.0003649417,0.0001899829,0.001183663,0.001343268,0.00003145527],"category_scores_gemma":[0.0004631052,0.001110054,0.0006811005,0.0005431904,0.0001732952,0.00001023438,0.001613281,0.0005109862,0.0000106738],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001877617,"about_ca_system_score_gemma":0.001360159,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002010943,"about_ca_topic_score_gemma":0.00001454753,"domain_scores_codex":[0.9953829,0.0002598146,0.0008597209,0.002001234,0.0003983465,0.001097949],"domain_scores_gemma":[0.995131,0.0001138017,0.0006573583,0.002904531,0.0008336197,0.0003597496],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00007381147,0.00014447,0.01797109,0.0005043983,0.001367743,0.000008945559,0.000005806926,0.02772931,0.94209,0.0003784698,0.009703218,0.00002279693],"study_design_scores_gemma":[0.000969676,0.0002423019,0.1173976,0.001537713,0.001551954,5.334226e-8,0.000003745181,0.008692374,0.8104227,0.0000261946,0.05558132,0.003574317],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9284889,0.0145472,0.05247686,0.00009280197,0.002147543,0.001490201,0.0004803379,0.0002405928,0.00003558262],"genre_scores_gemma":[0.9683807,0.001013107,0.02583536,0.0004072953,0.003350683,0.0006959423,0.00001006664,0.0001680186,0.0001388021],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1316672,"threshold_uncertainty_score":0.9999532,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00980508437019819,"score_gpt":0.2261701165698031,"score_spread":0.2163650321996049,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}