Skip to main content

Table 1 Performance of different algorithms

From: Penalized likelihood for sparse contingency tables with an application to full-length cDNA libraries

 

MSS

NLS

RMSE

SPREAD

Penalty-based regularization methods:

ℓ1-regularization

69.7%

2.20

0.228

0.144

Level-ℓ1-regularization

89.7%

2.22

0.237

0.179

Relaxed ℓ1-regularization

82.2%

2.22

0.233

0.154

ℓ2-regularization

-

2.20

0.238

0.130

MCMC without model selection:

   σ2 = 2

-

2.32

0.747

0.401

   σ2 = 1

-

2.27

0.467

0.287

   σ2 = 1/2

-

2.24

0.294

0.201

MCMC with model selection:

σ2~Γ-1(2,3)

81.5%

2.23

0.294

0.231

   σ2 = 2

76.6%

2.25

0.431

0.342

   σ2 = 1

78.4%

2.24

0.331

0.265

   σ2 = 1/2

76.6%

2.23

0.281

0.225

MCMC with hierarchical model selection:

σ2~Γ-1(2,3)

84.1%

2.22

0.255

0.180

   σ2 = 2

80.6%

2.29

0.415

0.284

   σ2 = 1

83.4%

2.26

0.308

0.221

   σ2 = 1/2

83.4%

2.24

0.247

0.178

   σ21 = 1/10

86.3%

2.20

0.236

0.097

   σ2 = 1/100

69.7%

2.28

0.420

0.033

  1. Comparison of different methods to estimate the interaction strength vector β. MSS, NLS, RMSE and SPREAD are described in the Implementation section. The additional methods relaxed ℓ1-regularization and ℓ2-regularization listed in the Table are explained in the Results Section.
Back to article page

BMC Bioinformatics

ISSN: 1471-2105

Contact us