Skip to main content

Table 4 The average test performance of SSL-GCN models with various unlabeled data ratio (\(R_{u}\) in brackets) on the 12 prediction tasks in 5 repeated experiments. For comparison, the results of the SL-GCN models are shown

From: Chemical toxicity prediction based on semi-supervised learning and graph convolutional neural network

Tasks

SL-GCN

SSL-GCN (0.5)

SSL-GCN (1.0)

AUC

Std.

AUC

Std.

AUC

Std.

NR-AR-LBD

0.6783

0.0269

0.7417

0.0105

0.7333

0.0401

NR-AR

0.7157

0.0367

0.7550

0.0483

0.7858

0.0357

NR-AhR

0.8260

0.0055

0.8161

0.0121

0.8295

0.0129

NR-Aromatase

0.7092

0.0167

0.7202

0.0057

0.7306

0.0156

NR-ER-LBD

0.6340

0.0161

0.6623

0.0330

0.6794

0.0411

NR-ER

0.6899

0.0160

0.7188

0.0196

0.7114

0.0179

NR-PPAR-gamma

0.6753

0.0278

0.7267

0.0210

0.7614

0.0212

SR-ARE

0.7134

0.0137

0.7241

0.0065

0.7288

0.0063

SR-ATAD5

0.6850

0.0223

0.7119

0.0080

0.7061

0.0245

SR-HSE

0.7644

0.0096

0.7636

0.0239

0.7678

0.0080

SR-MMP

0.7988

0.0066

0.8120

0.0075

0.8035

0.0061

SR-p53

0.6970

0.0253

0.7291

0.0114

0.7401

0.0203

Tasks

SSL-GCN (2.0)

SSL-GCN (3.0)

SSL-GCN (4.0)

AUC

Std.

AUC

Std.

AUC

Std.

NR-AR-LBD

0.7647

0.0279

0.7377

0.0145

0.7477

0.0135

NR-AR

0.7512

0.0358

0.7412

0.0659

0.7967

0.0251

NR-AhR

0.8287

0.0072

0.8303

0.0055

0.8224

0.0090

NR-Aromatase

0.7232

0.0040

0.7287

0.0082

0.7337

0.0057

NR-ER-LBD

0.6772

0.0161

0.6662

0.0250

0.6870

0.0282

NR-ER

0.7039

0.0124

0.7113

0.0083

0.7166

0.0137

NR-PPAR-gamma

0.7491

0.0201

0.7429

0.0177

0.7456

0.0223

SR-ARE

0.7297

0.0080

0.7277

0.0067

0.7243

0.0114

SR-ATAD5

0.7096

0.0139

0.7175

0.0143

0.7077

0.0162

SR-HSE

0.7822

0.0097

0.7731

0.0098

0.7700

0.0066

SR-MMP

0.8100

0.0033

0.8031

0.0088

0.8081

0.0078

SR-p53

0.7518

0.0198

0.7359

0.0147

0.7434

0.0126

  1. The bold number denotes the best result among all SSL-GCN models with various unlabeled data ratio in the corresponding task