Fig. 6: Conditional independence network analysis reveals the dependency structure of baseline immune factors.

a Receiver Operating Characteristic (ROC) curve of classifying the uninfected participants (n = 16) versus those with sustained infection (n = 18) using regularised logistic regression across repeated cross validation (CV) runs (n = 100). The diagonal grey dashed line represents the performance of a random binary classifier. b Top predictive features ranked by selection frequency across CV runs (n = 100). Error bars represent the mean +/− standard deviation of feature selection frequencies and coefficients obtained from 100 randomly partitioned train-test datasets (orange), and from 100 null models per CV run by permutation of response labels (grey). Empirical one-sided p-values were computed by comparing observed values with permutation-based null distributions. *p < 0.05, **p < 0.01. c Conditional independence sub-network of baseline immune factors centered on 1023 the CCL13-DC axis (grey convex hull). Edge color reflects Spearman correlation coefficient.