Fig. 2: T-cell classification models.

A(a) Representative multiplexed immunofluorescent tissue images and segmentation masks that were used for the multi-marker classification workflow. For single marker classification, the immune markers were assessed individually (CD3, CD4, CD8, FOXP3, PD1) and each cell was classified as positive or negative for each marker. A(b) Marker combination (AE1, CD3, CD4, CD8, FOXP3, PD1) was used for multi-marker classification workflow and based on the marker co-staining each cell was assigned a cell subtype as shown in the table. The eight multi-marker T-cell subtypes were as follows: T helper cells (Th), T helper PD1 (ThPD1), T cytotoxic (Tc), T cytotoxic PD1+ (TcPD1), T regulatory (Treg), T regulatory PD1 positive (TregPD1), Epithelial and other (non-lymphocyte, non-epithelial) cells. A(c) Illustration of the resulting multi-marker classification for the nuclear mask. B Distribution of calculated values for % of Total for each T-cell subtype, per patient. Each value represents the average of 2–3 assessable cores, per patient. C Total counts from all patients grouped as epithelial-associated cells (red) and stromal cells (green) for each multi-marker T-cell subtype. D Correlation matrix showing the relationship between different T-cell subtypes (Spearman’s correlation coefficients). A color-coded correlation scale is provided: blue ellipses represent positive correlations, while darker color and narrower ellipses correspond to larger correlation coefficient magnitudes. E Heatmap showing separation and clustering of patients based on % T cells of Total cells in tumor cores. Clusters based on the Ward.D agglomerative clustering method with Euclidean correlation distance measure.