Table 2 Performance of six tools on three biological tasks

From: Biological relevance of computationally predicted pathogenicity of noncoding variants

 Method

Task 1a

 

Task 2: Positional diagnosisb

Task 3: Diagnosis with noisy backgroundc

  

Adjd

UC

WC

LC

Pro

5'U

ups

int

UC

WC

LC

Pro

5'U

ups

int

CADD

o

>10 bp

+++

++

+

++

+

++

+++

++

+

−

+

+

o

+

CATO

–

>1 kbp

+

+

+

+

+

+

+

−

−

−

−

−

−

−

DeepSEA

o

>10 bp

+++

+++

+

+++

++

+++

+++

+++

+

o

++

+

+

+

EIGEN

–

>100 bp

++

++

+

++

+

++

++

+

+

o

o

+

+

o

GWAVA

–

>50 bp

−

−

−

−

−

−

−

+

+

+

++

++

+

−

LINSIGHT

–

>10 bp

+++

++

+

+++

+

++

++

+++

+

−

++

+

+

+

  1. Pathogenic variants are defined by location in ultra-conserved (UC), well-conserved (WC), or least-conserved (LC) intervals, or by location in the promoter (pro), 5’UTR (5’U), upstream gene region (ups) or intron (int)
  2. a– indicates pathogenic scores are not specific to alleles at the same position, o indicates allele-specific scores but with low discriminative power
  3. bsuccess rates are indicated by: − (<50%), o (50–60%), + (60–80%), ++ (80–90%), +++ (>90%)
  4. cnon-pathogenic vs. pathogenic ratios are indicated by: − (<1:1), o (1–2:1), + (2–10:1), ++ (10–20:1), +++ (>20:1)
  5. dAdj refers to adjacency corresponding to the minimum distance between pairs of pathogenic and non-pathogenic variants for which significantly different scores are produced in Task 2