{"id":"W2100922775","doi":"10.1093/bioinformatics/btv157","title":"Test set bias affects reproducibility of gene signatures","year":2015,"lang":"en","type":"article","venue":"Bioinformatics","topic":"Gene expression and cancer classification","field":"Biochemistry, Genetics and Molecular Biology","cited_by":123,"is_retracted":false,"has_abstract":true,"ca_institutions":"Princess Margaret Cancer Centre; University of Toronto; University Health Network; Montreal Clinical Research Institute","funders":"National Institute of General Medical Sciences; National Cancer Institute; National Center for Advancing Translational Sciences; BC Cancer Agency","keywords":"Normalization (sociology); Computer science; Data set; Data mining; Set (abstract data type); Test set; Reproducibility; Sample size determination; Population; Test data; Statistics; Pattern recognition (psychology); Artificial intelligence; Mathematics; Medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08116712,0.001367828,0.001720745,0.001988192,0.00238597,0.004870358,0.003172259,0.001943049,0.00330984],"category_scores_gemma":[0.3225425,0.0006377353,0.001675677,0.003081793,0.005795214,0.002891401,0.004025779,0.003260603,0.002678828],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001987108,"about_ca_system_score_gemma":0.002312413,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00215281,"about_ca_topic_score_gemma":0.002109603,"domain_scores_codex":[0.9034208,0.05378648,0.007411394,0.01778965,0.0156626,0.001929115],"domain_scores_gemma":[0.5654068,0.3347229,0.01289858,0.06536394,0.01988842,0.001719357],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.006141499,0.000658864,0.4602524,0.002113556,0.003610582,0.001086824,0.002319531,0.06442279,0.06282516,0.01599567,0.03212433,0.3484488],"study_design_scores_gemma":[0.0006794132,0.002182598,0.2959075,0.0005076925,0.001305177,0.002990347,0.001294027,0.2954113,0.2435668,0.1217313,0.03401235,0.0004115648],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.4617282,0.00338188,0.4950958,0.005236273,0.001893513,0.0008295311,0.007354883,0.01019973,0.01428008],"genre_scores_gemma":[0.9439228,0.0001744022,0.04390522,0.00137749,0.000235161,0.0005967413,0.006529426,0.001600431,0.001658293],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9188329,"threshold_uncertainty_score":0.4292579,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0516341996318329,"score_gpt":0.2907679073513346,"score_spread":0.2391337077195017,"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."}}