Fig. 7
Group survival association and combinations. a A Cox proportional hazard model is used to estimate relative risk for patients according to individual factor label values. Using this model, a relative risk of death can be calculated for each patient based upon her group label value. When patients are stratified into thirds according to relative risk, some factors show an association with survival indicated by the difference in Kaplan-Meier survival curves for the extremal groups (p-values shown in the graphs). The figure shows survival curves for two such factors, G56 and G173. The bottom 30%, middle 40%, and top 30% in terms of predicted survival under the coxph model are plotted in magenta, blue, and red, respectively. When these two factors are combined into a single risk model, the survival differential increases dramatically. b Single factor combinations with various Gene Ontology annotations can be combined and show synergy relative to survival association. The false discovery rate under a randomization test for six such pairs is below about 1/3 compared to only one single factor achieving that level. Note that it is not suggested that factors be selected based solely upon survival associations but rather factors that are interesting for other reasons, e.g. they contain druggable targets, can be investigated for an impact on survival both alone and in combinations