Figure 2.

Predictive accuracies of our custom XAI models in terms of their ability to predict the likelihood of ≥5-year survival of breast cancer patients based on estimated immune cell composition from bulk RNA-seq data using EPIC (a), CIBERSORT (b), TIMER (c), and xCell (d) cell type quantification methods. The XAI indicates that the B cells (e), M0 macrophages (f), CD8⁺ T cells (g), and NK T cells (h) (estimated using EPIC, CIBERSORT, TIMER, and xCell methods, respectively) are the most important immune cells in the TME features in predicting survivability of breast cancer patients. The features on the y-axis are used in the respective models; their relative positions were determined by their relative importance in making correct predictions. The blue dots represent lower feature values, and the red dots represent higher feature values. In these analyses, Shapley values < 0 represent “likely to survive longer than 5 years after diagnosis” while Shapley values > 0 represent “likely to die within 5 years” after diagnosis.