Skip to main content

Table 2 Performance of different graph reconstruction methods in simulations

From: Tailored graphical lasso for data integration in gene network reconstruction

Case

Edge disagreement %

Partial cor

Prior partial cor

Method

\(k_{\text{opt}}\)

Sparsity

Precision

Recall

1

0

0.2

0.2

Glasso

0.035

0.283

0.493

    

Wglasso

0.032

0.312

0.503

    

TailoredGlasso

49.64

0.031

0.389

0.606

2

0

0.2

0.1

Glasso

0.035

0.285

0.499

    

Wglasso

0.034

0.293

0.493

    

TailoredGlasso

13.39

0.034

0.295

0.496

3

0

0.1

0.2

Glasso

0.022

0.079

0.085

    

Wglasso

0.021

0.096

0.099

    

TailoredGlasso

3.34

0.021

0.100

0.103

4

0

0.1

0.1

Glasso

0.022

0.079

0.085

    

Wglasso

0.020

0.082

0.083

    

TailoredGlasso

5.63

0.020

0.083

0.084

5

10

0.2

0.2

Glasso

0.035

0.283

0.493

    

Wglasso

0.033

0.305

0.493

    

TailoredGlasso

41.2

0.032

0.335

0.532

6

20

0.2

0.2

Glasso

0.035

0.283

0.493

    

Wglasso

0.034

0.291

0.491

    

TailoredGlasso

4.31

0.034

0.291

0.493

7

100

0.2

0.2

Glasso

0.035

0.283

0.493

    

Wglasso

0.034

0.289

0.485

    

TailoredGlasso

2.94

0.034

0.290

0.485

  1. The performance of the different graph reconstruction methods in simulations. The edge disagreement between the graph of interest and its prior, as well as the size of the partial correlations in them, is shown as well. The results are averaged over \(N=100\) simulations. The best values of the different performance measures are marked in bold, and \(k_{\text{opt}}\) is the mean value of the k chosen by the eBIC selection criterion in the tailored graphical lasso (TailoredGlasso). The graphical lasso and weighted graphical lasso are abbreviated as Glasso and Wglasso, respectively