Skip to main content

Table 3 Prediction performances of available algorithms compared to pkaPS.

From: pkaPS: prediction of protein kinase A phosphorylation sites with the simplified kinase-substrate binding model

 

Prediction performance

 

Algorithm

S n [%]

S p [%]

Reference

DISPHOS

ca. 76

ca. 85

Iakoucheva et al. 2004 [27]

SCANSITE

70.7

92.9

Zhou et al. 2004 [15]

NETPHOSK

79

89

Blom et al. 2004 [12]

GPS

88.9

90.6

Xue et al. 2005 [16]

PREDPHOSPHO

88.3

91.1

Kim et al. 2004 [14]

pkaPS

95.8

93.5

this work

  1. The table shows the prediction performances of five other programs compared to the worst performance (S n from neighbor jackknife-test) of the pkaPS predictor. All listed values except for those from DISPHOS refer to PKA-specific versions of the prediction tools. The pkaPS program outperforms all currently available methods that can be used to detect PKA-dependent phosphorylation. The sensitivities (S n ) and specificities (S p ) were directly taken from the original papers. Iakoucheva et al. [27] provide two possible values for the specificity as performance measure for the DISPHOS predictor, one that takes into account the possible occurrence of noise in the negative learning set (higher S p ) and one that does not (lower S p ). For DISPHOS, the S n value of 76% for serine (none is mentioned for threonine) and the higher, estimated specificity value of 85% (for serine) were used. The prediction performance for the SCANSITE program was not taken from the corresponding publication (Obenauer et al. 2003 [13]) but from Table 2 in the GPS paper [15]. The reason is that no evaluation of the SCANSITE performance in detecting sites for phosphorylation by PKA could be found in the paper from Obenauer et al. 2003 [13]. Here, the values for the low stringency cut-off were taken.
Back to article page

Biology Direct

ISSN: 1745-6150

Contact us