Skip to main content

Table 4  Classification results on the MALDI MS data

From: Joint deep learning for batch effect removal and classification toward MALDI MS based metabolomics

 

Source

Target

Before calibration

After calibration

1

2

3

1

2

3

Sample

ACC

1

0.926

0.753

0.813

0.926

0.889

0.879

 

2

0.799

0.911

0.828

0.875

0.911

0.870

 

3

0.876

0.763

0.927

0.907

0.884

0.927

F-score

1

0.915

0.807

0.828

0.915

0.904

0.875

 

2

0.814

0.919

0.823

0.867

0.919

0.863

 

3

0.865

0.741

0.929

0.904

0.899

0.929

AUC

1

0.923

0.729

0.813

0.923

0.882

0.879

 

2

0.809

0.912

0.828

0.875

0.912

0.870

 

3

0.874

0.786

0.927

0.909

0.879

0.927

MCC

1

0.857

0.505

0.637

0.857

0.774

0.758

 

2

0.648

0.831

0.656

0.749

0.831

0.743

 

3

0.750

0.593

0.866

0.816

0.764

0.866

Subject

ACC

1

0.922

0.769

0.822

0.922

0.896

0.892

 

2

0.791

0.909

0.832

0 .891

0.909

0.876

 

3

0.871

0.769

0.925

0.921

0.892

0.925

F-score

1

0.911

0.821

0.837

0.911

0.910

0.890

 

2

0.814

0.918

0.827

0.883

0.918

0.870

 

3

0.862

0.749

0.926

0.916

0.905

0.926

AUC

1

0.919

0.744

0.822

0.919

0.889

0.892

 

2

0.802

0.909

0.832

0.890

0.909

0.875

 

3

0.870

0.793

0.925

0.921

0.888

0.925

MCC

1

0.848

0.539

0.658

0.848

0.789

0.784

 

2

0.636

0.827

0.666

0.780

0.827

0.753

 

3

0.740

0.608

0.858

0.841

0.779

0.858

  1. The top half is conducted in the sample level, and the bottom half in the subject level. When the source and target IDs are the same, we perform in-batch cross-validation, whose results are free of batch effect