{"id":"W4403508643","doi":"10.1107/s2052520624008679","title":"The seventh blind test of crystal structure prediction: structure ranking methods","year":2024,"lang":"en","type":"article","venue":"Acta Crystallographica Section B Structural Science Crystal Engineering and Materials","topic":"X-ray Diffraction in Crystallography","field":"Materials Science","cited_by":70,"is_retracted":false,"has_abstract":true,"ca_institutions":"Dalhousie University","funders":"Army Research Laboratory; Army Research Office; Division of Chemistry; Division of Materials Research; Japan Society for the Promotion of Science; Hrvatska Zaklada za Znanost; Science and Technology Facilities Council; Research Institute for Information Technology, Kyushu University; Natural Sciences and Engineering Research Council of Canada; Argonne National Laboratory; Oak Ridge Institute for Science and Education; Office of Science; Bill and Carol Fox Center for Humanistic Inquiry, Emory University; Fundación para el Fomento en Asturias de la Investigación Científica Aplicada y la Tecnología; Russian Science Foundation; University of Southampton; Workforce Development for Teachers and Scientists; Grantová Agentura České Republiky; XtalPi; Khalifa University of Science, Technology and Research; Dell EMC; Karl-Franzens-Universität Graz; Deutsche Forschungsgemeinschaft; Engineering and Physical Sciences Research Council; Agencia Estatal de Investigación; Oak Ridge Associated Universities; Eli Lilly and Company; U.S. Department of Energy; European Commission; University of Reading; National Science Foundation","keywords":"Ranking (information retrieval); Crystal structure prediction; Test (biology); Artificial intelligence; Computer science; Crystal structure; Statistics; Mathematics; Crystallography; Geology; Chemistry","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","sts","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.001831992,0.0006072838,0.0005682068,0.0008965233,0.001322954,0.001750866,0.0007971665,0.0003240546,0.0004154517],"category_scores_gemma":[0.0006100964,0.0004316765,0.0001850431,0.002481917,0.001642253,0.001619503,0.000335506,0.0005145631,2.44318e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001239114,"about_ca_system_score_gemma":0.0001865304,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008309902,"about_ca_topic_score_gemma":0.00002284807,"domain_scores_codex":[0.9956713,0.0001834951,0.001049963,0.00111657,0.00102482,0.0009538862],"domain_scores_gemma":[0.9975965,0.0007828932,0.0003486132,0.0006382471,0.0003250746,0.0003086509],"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":[0.00007309444,0.000008671758,0.0001064201,0.0002668375,0.00005356519,0.000004643938,0.000339206,0.0008931067,0.997067,0.0007413105,0.00006547642,0.0003806891],"study_design_scores_gemma":[0.001822627,0.000909605,0.041915,0.0007415653,0.0004196144,0.002452108,0.00124549,0.01198349,0.8849813,0.01050627,0.04087937,0.002143524],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9870979,0.0005957938,0.002472069,0.0001847082,0.007573316,0.0005697869,0.0008787825,0.0005846031,0.00004308524],"genre_scores_gemma":[0.9926475,0.0001334642,0.006330431,0.00001848241,0.0007025146,0.0000325677,0.00003703171,0.00006709703,0.00003094735],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1120856,"threshold_uncertainty_score":0.9999772,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00776343137360859,"score_gpt":0.2615333864569621,"score_spread":0.2537699550833535,"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."}}