Fig. 1: Overview of the universal immunohistochemistry (UIHC) artificial intelligence (AI) model development.

Single-cohort-derived models (SC-models) were trained using one dataset, while multiple-cohort-derived models (MC-models) were trained using multiple datasets, including lung, urothelial carcinoma, and breast cancer samples stained with Programmed Death-Ligand 1 (PD-L1) 22C3, as well as breast cancer samples stained with Human Epidermal growth factor Receptor 2 (HER2). The AI models’ performance was validated on both the training cohorts and novel cohorts that were not included in the training phase. These novel cohorts consisted of samples stained for Claudin 18.2, DeLta-Like 3 (DLL3), E-Cadherin, Fibroblast Growth Factor Receptor 2 (FGFR2), Human Epidermal growth factor Receptor 3 (HER3), Mesenchymal-Epithelial Transition factor (MET), MUcin-16 (MUC16), PD-L1 SP142, PD-L1 SP263, and TROPhoblast cell-surface antigen 2 (TROP2).