Table 1 Performances on a testing set of a vanilla neural network, kNN, linear models with different chemical features, TTWOPT and DeepCE with its simpler variants trained with different training sets

From: A deep learning framework for high-throughput mechanism-driven phenotype compound screening and its application to COVID-19 drug repurposing

Training sets

Models

Features

Pearson

r.m.s.e.

GSEA

Positive P@100

Negative P@100

Original

Vanilla neural network

PubChem

0.1101

  

ECFP

0.0705

  

Drug-target

0.1076

  

LTIP

0.0770

 

kNN

PubChem

0.0844

  

ECFP

0.1469

  

Drug-target

0.1811

  

LTIP

0.1231

High-quality

Vanilla neural network

PubChem

0.3929

1.8413

0.3853

0.2230

0.2622

  

ECFP

0.4105

1.8218

0.4049

0.2353

0.2690

  

Drug-target

0.4270

1.8002

0.4098

0.2334

0.2788

  

LTIP

0.4259

1.7843

0.4168

0.2361

0.2798

  

Random

0.3129

1.9152

0.3299

0.1729

0.2284

 

kNN

PubChem

0.3903

1.8464

0.3877

0.2089

0.2606

  

ECFP

0.3991

1.8264

0.4041

0.2186

0.2639

  

Drug-target

0.3907

1.8375

0.4105

0.2182

0.2625

  

LTIP

0.3922

1.8388

0.3959

0.2176

0.2578

 

Linear Regression

PubChem

0.1762

1.9821

0.2184

0.1220

0.1956

  

ECFP

0.1770

1.9916

0.2227

0.1232

0.1956

  

Drug-target

0.1763

1.9768

0.2216

0.1240

0.1957

  

LTIP

0.1764

1.9769

0.2232

0.1230

0.1956

 

Lasso

PubChem

0.1761

1.9775

0.2160

0.1203

0.1935

  

ECFP

0.1770

1.9763

0.2237

0.1198

0.1961

  

Drug-target

0.1764

1.9764

0.2177

0.1209

0.1935

  

LTIP

0.1764

1.9764

0.2177

0.1213

0.1916

 

Ridge Regression

PubChem

0.1762

1.9809

0.2185

0.1220

0.1961

  

ECFP

0.1770

1.9839

0.2254

0.1236

0.1953

  

Drug-target

0.1764

1.9764

0.2221

0.1232

0.1956

  

LTIP

0.1764

1.9762

0.2237

0.1215

0.1953

 

TT-WOPT

N/A

0.0133

1.9695

0.0121

0.1228

0.1342

 

Deep CE−attn

Neural FP

0.4418

1.7738

0.4088

0.2435

0.2827

 

Deep CE−drug−gene attn

 

0.4620

1.7418

0.4493

0.2667

0.3088

 

Deep CE−gene−gene attn

 

0.4477

1.7711

0.4244

0.2784

0.2961

 

Deep CE

 

0.4907

1.6889

0.4656

0.2885

0.3195

Augmented

Vanilla neural network

PubChem

0.4204

1.8140

0.3932

0.2282

0.2736

  

ECFP

0.4177

1.8102

0.4171

0.2191

0.2783

  

Drug-target

0.4302

1.8092

0.4263

0.2130

0.2785

  

LTIP

0.4299

1.7819

0.4237

0.2259

0.2810

 

kNN

PubChem

0.3973

1.8392

0.3927

0.2023

0.2615

  

ECFP

0.4121

1.8020

0.4204

0.2202

0.2809

  

Drug-target

0.4023

1.8072

0.4011

0.2232

0.2794

  

LTIP

0.4016

1.8223

0.3924

0.2184

0.2650

 

DeepCE

Neural FP

0.5014

1.6810

0.4735

0.2940

0.3249