Figure 8.

iEN machine learning identifies factors underlying tumor response to Ad-SF7-Fc

(A) Schematic overview of experimental design. (B) The receiver operating curve (ROC) of iEN model performance with and without inclusion of priors. (C) Comparison of iEN predicted values (arbitrary units) between responders, with statistical comparison performed with Wilcoxon sum rank test. (D) The iEN model network where each node is a feature entered into the model and is scaled by its individual area under the ROC (AUROC). Edges are significant spearman correlations between features (p < 0.05) after Bonferroni correction for multiple comparisons. Nodes are colored by the dataset they originate from and communities of shared immunological archetypes are highlighted. (E) The same network from (D), but with nodes colored by iEN model coefficients identifying features increased or decreased in responders and which contributed to model performance. Noncontributory features have a 0 iEN model coefficient. Data analyzed and presented are aggregated from two independent B16 tumor experiments.