{"id":"W4407574454","doi":"10.1021/acs.jcim.4c02296","title":"The Need for Continuing Blinded Pose- and Activity Prediction Benchmarks","year":2025,"lang":"en","type":"review","venue":"Journal of Chemical Information and Modeling","topic":"Stroke Rehabilitation and Recovery","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"Structural Genomics Consortium; University of Toronto","funders":"Structural Genomics Consortium; Canadian Institutes of Health Research; National Institute of General Medical Sciences; Strategic Innovation Fund; Genome Canada","keywords":"Computer science; Artificial intelligence; Machine learning","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":[],"consensus_categories":[],"category_scores_codex":[0.1511267,0.002026255,0.006386218,0.003157848,0.0009656028,0.006304068,0.009534753,0.003723316,0.004706835],"category_scores_gemma":[0.1563578,0.00100405,0.002857171,0.003409975,0.003205962,0.009522139,0.005774484,0.008922146,0.002793034],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002459993,"about_ca_system_score_gemma":0.01202503,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003012737,"about_ca_topic_score_gemma":0.003396267,"domain_scores_codex":[0.9247987,0.04078042,0.005772488,0.004945296,0.02233658,0.001366457],"domain_scores_gemma":[0.8485168,0.07260606,0.01027459,0.01878661,0.0467268,0.003089089],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0007556605,0.0007379255,0.003201475,0.0215947,0.001210822,0.0001205304,0.0005451528,0.01414389,0.004379657,0.04532482,0.0766075,0.8313777],"study_design_scores_gemma":[0.0007908793,0.002611274,0.009747163,0.03155811,0.001309446,0.0008023045,0.0009836337,0.01948828,0.01425201,0.09715056,0.8208969,0.000409339],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.01646122,0.6693743,0.2293858,0.03788868,0.01147716,0.003167587,0.005412508,0.002252169,0.0245806],"genre_scores_gemma":[0.1725539,0.3435307,0.4067966,0.02974724,0.006959193,0.01040348,0.02196805,0.002512458,0.005528377],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.1511267,"threshold_uncertainty_score":0.7992438,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03364179522404963,"score_gpt":0.3411454316147309,"score_spread":0.3075036363906812,"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."}}