{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00201907,0.001149936,0.0006589608,0.0007469829,0.0004252892,0.001577593,0.00292092,0.001309387,0.004461924],"category_scores_gemma":[0.01030547,0.0004941994,0.001364283,0.0004348902,0.001113315,0.002386016,0.001890256,0.002462484,0.001596654],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001455211,"about_ca_system_score_gemma":0.00212431,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007851654,"about_ca_topic_score_gemma":0.009754635,"domain_scores_codex":[0.9987304,0.0003005321,0.00006727293,0.0005331449,0.0002601256,0.0001085912],"domain_scores_gemma":[0.9957273,0.002870911,0.0002765426,0.0005539591,0.0004069975,0.0001642162],"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":[0.0008324501,0.0006694766,0.01372964,0.001050992,0.0002629583,0.000495907,0.0004533177,0.6110136,0.04578364,0.02063749,0.0227172,0.2823533],"study_design_scores_gemma":[0.00002692159,0.00004177005,0.0004598171,0.00002033172,0.00001666627,0.00003192226,0.00003101885,0.9775053,0.006787295,0.0135799,0.0014846,0.00001454527],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1599687,0.0008492409,0.7858463,0.001719735,0.0002581312,0.0002017403,0.004914078,0.03972361,0.006518408],"genre_scores_gemma":[0.7124575,0.0002253571,0.2743643,0.001221965,0.00006757342,0.0002203297,0.008100614,0.0007258975,0.002616379],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.007851654,"threshold_uncertainty_score":0.01561195,"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."}}