{"id":"W4415900298","doi":"10.1136/jitc-2025-sitc2025.0097","title":"97 Statistical and systematic uncertainties affect biomarker reliability: a case study in multiplex immunofluorescence","year":2025,"lang":"","type":"article","venue":"Regular and Young Investigator Award Abstracts","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Queen's University","funders":"","keywords":"Multiplex; Biomarker; Immunofluorescence; Affect (linguistics)","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0695259,0.0005897869,0.0010473,0.001678359,0.001811269,0.003089223,0.001995228,0.003145894,0.001317838],"category_scores_gemma":[0.2546202,0.0006763935,0.001536359,0.002093664,0.002837423,0.001939046,0.001459782,0.002345226,0.0002976884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0026654,"about_ca_system_score_gemma":0.002424189,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007637489,"about_ca_topic_score_gemma":0.005959797,"domain_scores_codex":[0.9616843,0.02369821,0.001928179,0.002891989,0.0086306,0.00116679],"domain_scores_gemma":[0.3944331,0.5604943,0.01544141,0.01216882,0.01644689,0.00101539],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002568738,0.0007804587,0.6043283,0.0007745561,0.001518563,0.02354442,0.009267254,0.1471733,0.01211623,0.01897945,0.004421872,0.1745269],"study_design_scores_gemma":[0.0003366755,0.003160324,0.2590333,0.0006153132,0.002014916,0.03042217,0.01030448,0.5459718,0.04559229,0.08881216,0.01297588,0.0007608496],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8545428,0.001411107,0.134347,0.003666146,0.00008733787,0.0002015144,0.0002238051,0.0003227176,0.005197569],"genre_scores_gemma":[0.9847878,0.0001293774,0.01445434,0.0001750485,0.00002506319,0.00002936994,0.0000467545,0.00005953203,0.0002927878],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9304741,"threshold_uncertainty_score":0.3676925,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06942844363114892,"score_gpt":0.3442540651931056,"score_spread":0.2748256215619567,"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."}}