A Multi-Method Approach for Proteomic Network Inference in 11 Human Cancers
Fig 2
Precision–recall (PR) curves optimized for full versus limited range of recall values.
(A) Top panel: PR curves for the 13 methods in the BRCA and GBM cohorts. PR curves are constructed by cumulatively increasing the number of edges from a ranked edge list. For each method, the relevant curve is computed with a choice of parameters that maximize AUPR in the recall range [0,1] (i.e. full-recall). Bottom panel: A zoomed-in version for recall in [0,0.1] and precision in [0,0.5]. (B) PR curves when the parameters are chosen to optimize AUPR specifically in the [0,0.1] recall range (i.e. limited-recall). We choose the limited-recall case for subsequent analysis because of two reasons. Beyond the 10% recall level, (1) the difference among methods become indiscernible, and (2) the precision level is very low suggesting network predictions are more likely to be affected by noise.