{"id":"W4409715624","doi":"10.1101/2025.04.21.649858","title":"A flaw in using pre-trained pLLMs in protein-protein interaction inference models","year":2025,"lang":"en","type":"preprint","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Bioinformatics and Genomic Networks","field":"Biochemistry, Genetics and Molecular Biology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Research in Immunology and Cancer; McGill University; Genome Canada; Mila - Quebec Artificial Intelligence Institute","funders":"Natural Sciences and Engineering Research Council of Canada; Alliance de recherche numérique du Canada","keywords":"Inference; Computer science; Protein–protein interaction; Artificial intelligence; Computational biology; Biology; Genetics","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":[0.01386269,0.001171043,0.001288609,0.0006965416,0.0007156385,0.001400676,0.003154163,0.001704867,0.001823417],"category_scores_gemma":[0.04952372,0.001216173,0.001258324,0.0007794682,0.001688655,0.004480138,0.003031388,0.004953499,0.001439646],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001516553,"about_ca_system_score_gemma":0.002022357,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004534689,"about_ca_topic_score_gemma":0.006884368,"domain_scores_codex":[0.9936517,0.003506964,0.0006514988,0.001204966,0.0007185635,0.0002662842],"domain_scores_gemma":[0.9679188,0.02201913,0.0009987699,0.006660248,0.001984784,0.0004183331],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002554428,0.0005151022,0.02755941,0.0007057463,0.001031421,0.0005135456,0.0004067188,0.6656584,0.01834049,0.008330074,0.0109962,0.2633885],"study_design_scores_gemma":[0.00004622425,0.0001671931,0.001497762,0.00004138776,0.00004678858,0.00009401875,0.00004270957,0.9794397,0.008500198,0.009008258,0.001091302,0.00002440907],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2773548,0.002420367,0.699504,0.003550295,0.0003864245,0.0002019828,0.001792489,0.01216775,0.002621942],"genre_scores_gemma":[0.8732619,0.0003036766,0.1204465,0.001525921,0.0001047389,0.0002079313,0.002292702,0.0005082654,0.00134839],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9861373,"threshold_uncertainty_score":0.07331377,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01627547318719987,"score_gpt":0.2503000072775701,"score_spread":0.2340245340903702,"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."}}