Table 5

(i) Impact of introducing different levels of artificial noise on MCL and MCL-CAw (ii) Role of affinity scoring in reducing the impact of noise

Method

PPI

Network

#Predicted complexes

#Matched predictions

Precisions

#Derivable benchmarks

#Derived benchmarks

Recall


MCL

G+K

242

55

0.226

182

62

0.338

G+K+Rand2k

265

56

0.215

182

64

0.352

G+K+Rand5k

274

61

0.223

182

68

0.379

G+K+Rand10k

316

64

0.202

182

69

0.379

ICD(G+K)

119

73

0.613

153

73

0.477

ICD(G+K+Rand2k)

104

59

0.567

153

66

0.431

ICD(G+K+Rand5k)

108

60

0.546

151

65

0.430

ICD(G+K+Rand10k)

112

60

0.546

150

65

0.433


MCL-CAw

G+K

310

77

0.248

182

77

0.423

G+K+Rand2k

140

59

0.421

182

68

0.374

G+K+Rand5k

116

62

0.534

182

70

0.384

G+K+Rand10k

176

64

0.363

182

68

0.373

ICD(G+K)

129

80

0.620

153

80

0.523

ICD(G+K+Rand2k)

102

62

0.608

153

73

0.477

ICD(G+K+Rand5k)

102

64

0.627

151

76

0.503

ICD(G+K+Rand10k)

106

64

0.603

150

76

0.506


The Gavin+Krogan network was introduced with 2000 - 10000 (10% to 75%) random interactions. Following this, these noisy networks were scored using the ICD scheme. With the aid of scoring, MCL-CAw was able to perform better than MCL even at 50% random noise.

Srihari et al. BMC Bioinformatics 2010 11:504   doi:10.1186/1471-2105-11-504

Open Data