Skip to main content

Table 9 Performance benchmark with the LD40 dataset under parameters at comparable high specificity level.

From: Parameterization of disorder predictors for large-scale applications requiring high specificity by using an extended benchmark dataset

Method

threshold

sensitivity

specificity

MCC

PE

IUPred long

0.54

0.597

0.948

0.602

0.545

DISOPRED2

0.08

0.587

0.947

0.591

0.534

IUPred short

0.51

0.521

0.948

0.539

0.469

CAST

40

0.510

0.951

0.535

0.462

SEG45

3.30;3.60

0.419

0.950

0.454

0.369

DisEMBL Rem465

1

0.359

0.969

0.437

0.328

SEG25

2.94;3.24

0.373

0.946

0.406

0.319

SEG12

2.29;2.59

0.291

0.950

0.333

0.240

DisEMBL Coils

1.94

0.263

0.948

0.302

0.211

DisEMBL Hotloops

2.7

0.262

0.949

0.301

0.210

  1. Predictors were run and ranked as in Tables 3 and 4 with parameters tuned to produce comparable high specificity of ~0.950 (~5% of false positive predictions) but over the LD40 dataset which only includes long disordered regions (length of 40 residues and above). For DisEMBL Remark 465 (in italic), parameter tuning only allowed a specificity of 0.969 as closest value to our criterion.