Fig. 2: AI-guided optimization of microbial product cocktails.

A Workflow for AI-guided MPC optimization. B Heatmap illustrating NK cell (NK-92MI) recruitment by UM-UC-1 spheroids in response to individual microbial products. C Heatmap comparing NK cell recruitment with combinations of microbial products, illustrating interactions between microbial products. D Schematic of the neural network model used to predict immunorecruitment activities of MPCs, generated using NN-SVG (MIT License; associated publication licensed under CC BY 4.0). E Heatmap of the 60 highest output combinations, showcasing the most effective MPCs as predicted by the neural network model, and a heatmap of the 60 lowest output combinations, highlighting those MPCs with lesser effectiveness as predicted by the model. NK cell recruitment-enhancing capabilities of MPC1-12 on UM-UC-1 cells tested in (F) the 16-well microwell device and (G) the 4-well format for selected MPCs. NK cell recruitment-enhancing capabilities of MPC1-12 on primary patient cancer cells (H) DT2101, (I) DT2334, (J) DT2153, and (K) DT2196. These samples cover MIBC from stage T2 to stage T4. Data are presented as mean ± SE. P values were derived by the Welch T test, ns P > 0.05; *P < 0.05; **P < 0.01. n = 4, independent biological replicate experiments in (G) and n = 3, independent biological replicate experiments in all other experiments. Source data are provided as a Source Data file.