Skip to main content

Table 4 Precision and recall for unseen concepts in the test set

From: Parallel sequence tagging for concept recognition

Annotation set

Harmonisation

Proper

Extended

Ref. count

Prec.

Recall

Ref. count

Prec.

Recall

Unique

Occ.

  

Unique

Occ.

  

CHEBI

Spans-only/-first

110

447

0.7747

0.5199

134

538

0.6265

0.5462

Ids-first

  

0.8805

0.0867

  

0.7131

0.0530

CL

Spans-only/-first

52

484

0.8817

0.2222

52

484

0.6900

0.2338

Ids-first

  

0.8750

0.0723

  

0.4326

0.0152

GO_BP

Spans-only/-first

120

484

0.6170

0.1402

126

508

0.2699

0.1621

Ids-first

  

0.7466

0.0524

  

0.4243

0.0175

GO_CC

Spans-only/-first

32

184

0.6343

0.1965

36

231

0.5158

0.1853

Ids-first

  

0.4956

0.0458

  

0.3375

0.0058

GO_MF

Spans-only/-first

1

1

–

–

73

416

0.5366

0.1393

Ids-first

  

–

–

  

0.5329

0.0090

MOP

Spans-only/-first

2

2

–

–

2

2

–

–

Ids-first

  

–

–

  

–

–

NCBITaxon

Spans-only/-first

40

87

0.3805

0.2974

44

95

0.4070

0.3342

Ids-first

  

0.7346

0.1689

  

0.7363

0.1163

PR

Spans-only/-first

278

4782

0.8170

0.7350

309

5156

0.8286

0.7230

Ids-first

  

0.7402

0.1200

  

0.6852

0.0909

SO

Spans-only/-first

16

101

0.0962

0.0571

25

123

0.1586

0.2579

Ids-first

  

1.0000

0.0198

  

0.9345

0.1823

UBERON

Spans-only/-first

203

1297

0.7246

0.2447

207

1308

0.7142

0.2555

Ids-first

  

0.5913

0.0342

  

0.5342

0.0225

  1. For each annotation set, the number of annotations (ref. count) in the test set is given, counting both occurrences (occ.) and unique labels (unique). A dash for precision and recall means that the corresponding system did not predict any unseen concept at all (neither true nor false positive)