Table 2

Classification results

Array

Flagging

# probes

Method

Sens.

Spec.

PPV

NPV

S. aureus

Weak

4515

BCSP

0.968

0.938

0.987

0.861

Mix.mod.

0.945

0.901

0.979

0.776

p-value

4.1·10^-7

7.4·10^-4

1.7·10^-4

2.5·10^-7

Hard

3539

BCSP

0.977

0.864

0.989

0.754

Mix.mod.

0.960

0.825

0.985

0.607

p-value

3.2·10^-8

0.02

0.003

3.7·10^-8

E. faecalis

Weak

3228

BCSP

0.989

0.953

0.983

0.560

Mix.mod.

0.946

0.890

0.989

0.458

Hard

3145

BCSP

0.992

0.0

0.869

0.0

Mix.mod.

0.955

0.474

0.990

0.409

Classification of probes into Present/Absent was done after both weak and hard flagging, see text for details, and using both our bias-corrected S-signal prediction method (BCSP) and the gaussian mixture model method. The number of probes indicate the median number of ordinary probes in a data set after flagging. All the results shown (Sens., Spec., PPV and NPV) are averages over the 28 S. aureus or 4 E. faecalis data sets. The p-values listed for the S. aureus results are the outcome of a Wilcoxon signrank test of wether the BCSP-method is a significant improvement of the mixture model method. A smaller p-value indicate a significant improvement. For the E. faecalis results all p-values were large (> 0.10) and they are not listed.

Snipen et al. BMC Bioinformatics 2009 10:91   doi:10.1186/1471-2105-10-91