biostudies-literatureUnknown15(9)Cai ZArtificial intelligence applications in biomedicine face major challenges from data privacy requirements. To address this issue for clinically annotated tissue proteomic data, we developed a federated deep learning approach (ProCanFDL), training local models on simulated sites containing data from a pan-cancer cohort (n = 1,260) and 29 cohorts held behind private firewalls (n = 6,265), representing 19,930 replicate data-independent acquisition mass spectrometry runs. Local parameter updates were aggregated to build the global model, achieving a 43% performance gain on the hold-out test set (n = 625) in 14 cancer subtyping tasks compared with local models and matching centralized model performance. The approach's generalizability was demonstrated by retraining the global model with data froCancer discovery1803-1818https://www.ebi.ac.uk/biostudies/studies/S-EPMC12409279biostudies-literatureFederated Deep Learning Enables Cancer Subtyping by Proteomics.PMC12409279Horvath LGHarris CCGraham JDBoys ELFlynn PJMann GJArmengol CCooper WAKoh JMSPanizza BJLord RVBowman EDLucas NWilliams SGWu YConnolly EABasik MNagabushan SBarranco MMHabib RGill AJReddel RRDalal BMcCaughan GWZhong QHarvey KFilipits MAguilar-Mahecha AGnant MGeorge JHains PGLiu JJBalleine RLSamra JSSouglakos JThomas DKohonen-Corish MRJSwarbrick ANoor ZJackson CMurphy LAref ATDeFazio AMacKenzie KLClifton-Bligh RPoulos RCDausmann MRobinson PJLong GVLim ENavines JCai ZXavier DNagrial AScolyer RAMorgan LfalseFederated Deep Learning Enables Cancer Subtyping by Proteomics.Artificial intelligence applications in biomedicine face major challenges from data privacy requirements. To address this issue for clinically annotated tissue proteomic data, we developed a federated deep learning approach (ProCanFDL), training local models on simulated sites containing data from a pan-cancer cohort (n = 1,260) and 29 cohorts held behind private firewalls (n = 6,265), representing 19,930 replicate data-independent acquisition mass spectrometry runs. Local parameter updates were aggregated to build the global model, achieving a 43% performance gain on the hold-out test set (n = 625) in 14 cancer subtyping tasks compared with local models and matching centralized model performance. The approach's generalizability was demonstrated by retraining the global model with data fro2025-01-01T00:00:00Z2025 Sep2026-05-29T22:38:27.052Z2026-04-08T06:17:33.939ZS-EPMC124092794048862010.1158/2159-8290.CD-24-1488