{"id":"W4411597815","doi":"10.1101/2025.06.17.659642","title":"Biological Reasoning with Reinforcement Learning through Natural Language Enables Generalizable Zero-Shot Cell Type Annotations","year":2025,"lang":"en","type":"preprint","venue":"bioRxiv (Cold Spring Harbor Laboratory)","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Reinforcement learning; Zero (linguistics); Computer science; Reinforcement; Shot (pellet); One shot; Artificial intelligence; Natural (archaeology); Natural language; Psychology; Linguistics; Biology; Engineering; Chemistry; Social psychology","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":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0005713868,0.0007481732,0.0006524271,0.0002993416,0.000550615,0.0007960511,0.001764233,0.0004721738,0.00003415454],"category_scores_gemma":[0.0003272983,0.0006746102,0.0001465746,0.001262019,0.0001373848,0.0005394341,0.001678042,0.001619959,0.00006074389],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003727903,"about_ca_system_score_gemma":0.0009052285,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001741793,"about_ca_topic_score_gemma":0.000001395346,"domain_scores_codex":[0.9961061,0.000275156,0.0006736335,0.00133874,0.0006586355,0.0009477491],"domain_scores_gemma":[0.996596,0.0001592568,0.0006753406,0.00168761,0.000700239,0.0001815352],"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":[0.00002717445,0.00004795514,0.002300815,0.0002743175,0.0001591882,0.0000864284,0.0002166064,0.9485704,0.03981308,0.007957068,0.0005451508,0.000001831747],"study_design_scores_gemma":[0.001742468,0.0006953098,0.004376947,0.002021421,0.0002404046,2.193627e-7,0.00007440286,0.8182184,0.1441976,0.000008182443,0.02511048,0.003314216],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.05534274,0.002267571,0.9377534,0.0001588517,0.001572747,0.000852253,0.00001106954,0.001313164,0.0007282268],"genre_scores_gemma":[0.8276032,0.0003411257,0.1706366,0.0004136734,0.0001706808,0.0001001708,0.000004051865,0.00005727754,0.000673204],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7722605,"threshold_uncertainty_score":0.9995705,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01926935263878364,"score_gpt":0.2416836232015132,"score_spread":0.2224142705627296,"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."}}