,

yp

es were the OS and ORT models. The maximum Type I error rate

st 23% for the OS model for four outliers data, which resulted in

I errors.

e 6.7. Prediction errors of non-DEGs for the simulated data for all tests.

Outlier

1

2

3

4

5

T test

0

0

0

0

0

Limma

0

0

0

0

3

COPA

0

2

9

15

17

OS

78

78

88

91

84

ORT

46

42

51

54

61

MOST

8

16

21

17

17

LSOSS

0

0

0

0

0

DOG

0

0

0

0

0

e 6.8. A comparison of prediction error of DEGs for the simulated data.

Error of up-regulated DEGs

Error of down-regulated DEGs

1

2

3

4

5

1

2

3

4

5

0

2

7

8

10

6

10

10

10

10

0

0

6

8

10

4

10

10

10

10

2

1

1

1

2

50

50

50

48

40

0

0

0

0

0

43

42

37

37

37

4

3

3

3

2

38

34

37

36

38

39

37

36

33

37

49

48

48

50

47

31

36

36

39

37

50

50

50

50

50

0

0

0

0

4

2

2

3

3

2

6.8 shows the error statistics (Type II error) of all tests for the

gulated DEGs and the up-regulated DEGs. Except for the MOST

SOSS models, others had small errors for the up-regulated DEGs.

not a surprise for the COPA, OS and ORT models. These

ms were developed based on the similar working principle, i.e.,

the outliers were present in the case replicates only. The MOST

SS models showed unreasonable results because many DEGs

outliers were also mis-classified. In terms of the down-regulated

was not a surprise when the COPA, OS and ORT models returned

performance. As aforementioned, these three algorithms only