{"id":"W4394769874","doi":"10.1093/bioinformatics/btae196","title":"Effect of tokenization on transformers for biological sequences","year":2024,"lang":"en","type":"article","venue":"Bioinformatics","topic":"Machine Learning in Bioinformatics","field":"Biochemistry, Genetics and Molecular Biology","cited_by":41,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Tel Aviv University; Azrieli Foundation; Israel Science Foundation","keywords":"Lexical analysis; Computer science; Artificial intelligence; Natural language processing; Biological network; Inference; Security token; Machine learning; Computational biology; Biology","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004819325,0.0001492767,0.0001584768,0.00007657011,0.00004686324,0.00003243941,0.0001233546,0.0001651876,0.00001243461],"category_scores_gemma":[0.000235617,0.0001018225,0.0001319716,0.0001139944,0.00007814601,0.00001043494,0.00002078402,0.00007920546,0.00001923697],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001208364,"about_ca_system_score_gemma":0.00004320759,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000001120166,"about_ca_topic_score_gemma":5.676803e-7,"domain_scores_codex":[0.999221,0.00003063347,0.0003527811,0.00009998439,0.0001244749,0.000171101],"domain_scores_gemma":[0.9996098,0.0001141604,0.00007425448,0.0001184466,0.00003999357,0.00004338802],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.001860544,0.0001295188,0.00331801,0.01215369,0.00063975,0.00000282062,0.002387423,0.005503284,0.1039274,0.0116519,0.02244208,0.8359836],"study_design_scores_gemma":[0.001263536,0.01403649,0.0001617402,0.0002951892,0.00009400266,0.00003437633,0.0001951336,0.2947741,0.5435262,0.000106209,0.1449502,0.0005628031],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7306715,0.0004491503,0.2509752,0.0002148511,0.0008066813,0.001332777,0.0002166953,0.0001325456,0.01520063],"genre_scores_gemma":[0.9914115,0.0001569403,0.007494724,0.0001363296,0.0001120643,0.00003639006,0.0004813415,0.0000160827,0.000154617],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8354207,"threshold_uncertainty_score":0.4152203,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009661397653894933,"score_gpt":0.2905729843578977,"score_spread":0.2809115867040027,"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."}}