Fig. 5. Unsupervised integrative clustering of patients with CSM by similarity network fusion (SNF).

(A) Alluvial plot tracing clustering path of study subjects based on the combined set of clinical variables, metabolites, proteins, and cytokines quantified. (B) Patient similarity matrix and network based on the two and four SNF clusters and their associated network representation. In the patient-similarity network, each circle denotes an individual; edges connecting individuals are weighted by the similarity score and colored by contributing data type(s) as per legend. (C) Differences in time to death and time to discharge of subjects grouped into SNF clusters C, D, E, and F. Statistical test was performed by Cox proportional hazard analysis of event time. n.s., not significant. (D) Z-score scatterplots of metabolites, proteins, and inflammatory mediators of children who died during the first 3 days, during day 4 to day 7, and beyond 7 days of hospitalization. Zero on the x axis represents the mean of children who survived from the respective analyte. The higher the Z-score, the more deviated the analyte is in NS with respect to S.