Table 2 Classification performance of metrics based on relative or absolute gene expression levels.

From: Bipolar patients display stoichiometric imbalance of gene expression in post-mortem brain samples

 

Relative expression (sWSR)

Absolute expression

 

SI score

Predicted probabilities

Predicted probabilities

Dataset

Genes

p

AUC

R2

Genes

p

AUC

R2

Genes

p

AUC

R2

CMC-HBCC

55

2.5E-05

0.67

5.3%

54

1.2E-10

0.76

17.4%

54

9.9E-02

0.57

1.3%

CMC-Pitt

55

3.7E-02

0.62

1.8%

54

7.4E-04

0.69

10.0%

54

2.8E-02

0.63

3.2%

BrainGVEX-SMRI

54

7.8E-04

0.66

6.1%

54

1.7E-05

0.71

10.2%

54

4.0E-01

0.54

0.9%

BipSeq-sACC

55

1.4E-07

0.69

4.8%

54

3.4E-11

0.74

13.5%

54

7.4E-03

0.60

1.8%

  1. Predicted probabilities are estimated with validation across datasets: to predict the probabilities of a test dataset, the logistic regression model is fitted using the three other datasets and predicted probabilities are computed for the test dataset from the fitted model. This required removal of one gene that does not have expression in all datasets.
  2. p: p value from the Wilcoxon Rank Sum test (Mann-Whitney-Wilcoxon) of difference between cases and controls.
  3. AUC: Area under the ROC curve.
  4. R2: Nagelkerke pseudo-R2 (liability-adjusted) from the logistic regression of diagnosis as a function of the SI score or the predicted probability.
Back to article page