Table 3

Ranking of gene set methods

Rank

Methods

Accuracy

Sets

Rank. Algo

Aggrgt

Median

Avg

σ

Iqr


1

1:10

Global

SVD

89.2

79.5

18.9

33.2

2

1:10

Global

None

88.3

81.0

17.7

31.3

3

1

Global

None

87.8

80.7

17.5

31.0

4

1:10

Global

SetSig

87.4

81.1

16.5

26.1

5

1:10

Global

AVG

85.6

78.7

18.4

32.6

6

1:10

SAM-GS

SetSig

85.4

79.9

17.1

30.2

7

1:10

SAM-GS

None

84.6

80.1

17.3

30.7

8

1

Global

SVD

83.8

77.9

20.1

34.3

9

1:10

GSEA

SetSig

83.4

78.3

16.7

26.3

10

1:10

GSEA

None

82.3

80.0

16.8

30.4

11

1:10

SAM-GS

SVD

79.9

77.1

18.0

32.1

12

1:10

GSEA

SVD

79.2

77.2

17.7

31.7

13

1:10

GSEA

AVG

79.1

76.4

16.9

31.9

14

1

SAM-GS

None

78.3

76.0

15.3

26.3

15

1

Global

SetSig

77.5

75.9

15.1

23.5

16

1

GSEA

None

76.7

75.6

16.3

29.5

17

baseline (all genes used)

75.5

76.6

18.4

33.5

18

1

SAM-GS

SetSig

75.0

74.7

14.2

18.9

19

1

Global

AVG

72.7

73.8

17.6

31.1

20

1:10

SAM-GS

AVG

72.5

73.8

15.9

26.0

21

1

GSEA

SetSig

70.2

72.6

17.0

26.8

22

1

GSEA

AVG

69.6

68.1

12.8

22.4

23

1

GSEA

SVD

69.5

71.9

16.3

28.2

24

1

SAM-GS

SVD

69.0

69.5

15.7

21.3

25

1

SAM-GS

AVG

67.3

67.0

11.4

15.5


Ranking of combinations of gene set methods by median predictive accuracy achieved on 30 datasets (Table 8, Section Expression and gene sets) with 5 machine learning algorithms (Section Machine learning) estimated through 10-fold cross-validation (i.e. 1,500 experiments per row). The columns indicate, respectively, the resulting rank by median accuracy, the gene sets used to form features (1 - the top ranking set, 1:10 - the top ten ranking sets), the gene set selection method, the expression aggregation method (see Section Methods and data for details on the latter 3 factors), and the median, average, standard deviation and interquartile range of the accuracy.

Holec et al. BMC Bioinformatics 2012 13(Suppl 10):S15   doi:10.1186/1471-2105-13-S10-S15

Open Data