Skip to main content

Table 3 Performance of the prognostic models comprised of various 5hmC and transcriptomic features

From: Dissection of transcriptomic and epigenetic heterogeneity of grade 4 gliomas: implications for prognosis

Data

5hmC

Transcriptome

5hmC-Transcriptome

Feature type

Promoter

H3K27ac

Gene body

mRNA

Integrated

Model

Glmboost

Glmnet

Glmboost

Glmboost

Glmboost

Feature Selection Method

rf

uc

rf

rf

rf

Feature Number

30

20

20

10

30

IDH1

0.57 (0.47–0.67)

Age + Gender + IDH1

0.68 (0.53–0.83)

PS

0.74 (0.60–0.87)

0.72 (0.58–0.86)

0.69 (0.53–0.85)

0.70 (0.56–0.85)

0.74 (0.60–0.88)

Age + Gender + PS

0.74 (0.61–0.87)

0.70 (0.58–0.83)

0.70 (0.57–0.83)

0.72 (0.58–0.85)

0.71 (0.57–0.86)

Age + Gender + IDH1 + PS

0.72 (0.59–0.86)

0.69 (0.55–0.84)

0.68 (0.54–0.83)

0.72 (0.59–0.85)

0.72 (0.58–0.86)

  1. Average Harrell’s concordance index (c-index) and 95% confidence intervals (CI) of the testing sets are shown for each model
  2. glmboost gradient boosted generalized linear survival learner, glmnet generalized linear survival learner with the elastic net regularization, rf random forest, uc univariate Cox proportional hazards model, PS prognostic signatures based on 5hmC, transcriptome, or integrated, IDH1 isocitrate dehydrogenase 1