Skip to main content
Figure 4 | Biology Direct

Figure 4

From: Test on existence of histology subtype-specific prognostic signatures among early stage lung adenocarcinoma and squamous cell carcinoma patients using a Cox-model based filter

Figure 4

Sensitivity curves for different cutoffs in the Cox-model filter. A) Predictive statistics as a function of different FDR cutoff in moderated-t test filtering; B) Predictive statistics as a function of different FDR cutoff in Cox-model filter, using 6,202 genes. Here, No.1/No.2/No.3 represent the overall/AC/SCC size of selected gene sets, respectively; the number in the following line represents the total genes under consideration in Panel A; accuracy stands for the proportion of rightly classified patients; risk scores were constructed using PCA method since the number of selected genes is bigger than the number of samples for most values in Panel B. From these two plots, we observed that 0.05 is an optimal threshold in both steps based on the model parsimony and predictive performance.

Back to article page