Skip to main content

Table 1 Ranking results in the training phase based on the CircularFingerprint

From: Improving MetFrag with statistical learning of fragment annotations

Top1

Top3

Top10

Top1 (%)

α

β

αL

βL

ω1

ω2

ω3

# Spectra

Negative Mode

55

93

151

20.8

0.00002

0.00250

0.00050

0.00050

0.268

0.460

0.272

265

51

89

155

19.5

0.00002

0.06250

0.01250

0.00050

0.434

0.380

0.186

261

62

101

165

22.9

0.00050

0.01250

0.00010

0.01250

0.309

0.508

0.184

271

70

106

170

25.8

0.00050

0.00250

0.00002

0.01250

0.317

0.494

0.189

271

62

103

161

23.8

0.00010

0.00010

0.00010

0.00250

0.170

0.616

0.214

260

67

110

153

24.0

0.00010

0.00250

0.00250

0.00010

0.300

0.493

0.207

279

63

98

157

22.9

0.00010

0.00050

0.00010

0.00050

0.054

0.512

0.434

275

68

102

158

25.0

0.00002

0.00250

0.00250

0.00250

0.240

0.558

0.202

272

86

114

171

31.2*

0.00010

0.00250

0.00250

0.00010

0.413

0.398

0.189

276

74

106

161

29.0

0.00010

0.00010

0.00002

0.00010

0.189

0.465

0.346

255

Positive Mode

412

664

925

28.0

0.00010

0.00250

0.00010

0.00250

0.333

0.438

0.229

1471

402

622

866

28.2

0.00010

0.00050

0.00010

0.00250

0.208

0.483

0.309

1426

406

665

913

29.0

0.00010

0.01250

0.00250

0.00250

0.333

0.438

0.229

1399

395

651

894

27.6

0.00010

0.00250

0.00250

0.00250

0.309

0.503

0.188

1432

387

618

839

27.4

0.00010

0.00250

0.00050

0.00050

0.413

0.398

0.189

1413

408

630

870

28.6

0.00010

0.00050

0.00050

0.00050

0.165

0.584

0.251

1428

432

655

910

30.6*

0.00010

0.01250

0.00250

0.00050

0.378

0.488

0.134

1410

400

642

874

28.2

0.00010

0.00250

0.00250

0.00050

0.210

0.488

0.302

1420

385

613

830

27.7

0.00010

0.00250

0.00010

0.00010

0.266

0.388

0.346

1389

396

638

891

27.7

0.00010

0.00050

0.00050

0.00010

0.165

0.593

0.242

1428

  1. The optimization of the parameters was performed on the training data set with ten different random splits of the MS/MS training spectra to be used for first and second training phase. Optimization was performed separately for positive and negative mode. *Runs with the best results based on the relative correct Top1 rankings (neg: R09, pos: R07)