{"id":"W4415376838","doi":"10.1101/2025.10.20.683304","title":"Deep Learning-Based Genetic Perturbation Models <i>Do</i> Outperform Uninformative Baselines on Well-Calibrated Metrics","year":2025,"lang":"","type":"preprint","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Bioinformatics and Genomic Networks","field":"Biochemistry, Genetics and Molecular Biology","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Benchmarking; Deep learning; Metric (unit); Interpretability; Correlation; Calibration; Mean squared error; Baseline (sea)","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":[],"consensus_categories":[],"category_scores_codex":[0.01277853,0.002441394,0.001305382,0.001418791,0.0005411805,0.002430109,0.002059373,0.001780304,0.001172798],"category_scores_gemma":[0.02429121,0.0004460476,0.001136564,0.001390124,0.002045164,0.003809397,0.001760884,0.002921377,0.0009331519],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002645855,"about_ca_system_score_gemma":0.00185031,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005396719,"about_ca_topic_score_gemma":0.00625894,"domain_scores_codex":[0.9925897,0.002534187,0.0005124671,0.001973414,0.002008646,0.0003815093],"domain_scores_gemma":[0.9878438,0.004506343,0.001693496,0.00398738,0.001598595,0.0003704479],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0008349106,0.0003347354,0.0186236,0.0009260338,0.0009255161,0.000142097,0.0001007223,0.7618158,0.02716147,0.01617557,0.01359834,0.1593612],"study_design_scores_gemma":[0.0000405145,0.0005797883,0.005281725,0.0001394953,0.0001173257,0.0001223534,0.0000637402,0.9128676,0.04728667,0.02762714,0.005767308,0.0001063067],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3274098,0.006869468,0.6314935,0.003406632,0.0007368303,0.0002563999,0.005038912,0.01482655,0.009961854],"genre_scores_gemma":[0.8724307,0.001143665,0.1166689,0.001136186,0.00008250638,0.0001519771,0.005617158,0.000825013,0.001943951],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01277853,"threshold_uncertainty_score":0.06758016,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009366974811906304,"score_gpt":0.2037573142482187,"score_spread":0.1943903394363124,"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."}}