g

g

p

wo conditions. This kind of error is named as a Type I error. A

error, on the other hand, refers to a falsely accepted null

is. In the context of gene differential expression analysis, a Type

fers to a wrongly predicted DEG and a Type II error stands for a

predicted non-DEG, i.e., a missing DEG [Li, et al., 2013; Mudge,

17; Gonzalez, et al., 2020].

ype I and Type II errors often occur in an experiment especially

sample size of a data set is small. This is because it is often to

a few replicates to cut the expenses in many biological

nts. A severe outcome for a data set with insufficient replicate is

able population variance in addition to an unreliable population

letrozole drug data which employs 58 baseline replicates and 58

replicates, the gene EHD2 is a DEG and the gene STAT1 is a

G. Suppose a subset of replicates is randomly drawn from 58

for these two genes. It is interesting to observe how the

al expression status of these two genes change. For this

tion, the Type I and Type II errors were counted. The drawing

tarted from two till 50. For each drawn sample with a specific

number, 100 random simulations were carried out and hence 100

alues were calculated using the t test. Based on these 100 t test p

e Fisher test [Fisher, 1948] for combining p values was used. The

st statistic is defined as below,

ܨൌെ2 ෍lnሺ݌௞ሻ

௄

௞ୀଵ

(6.5)

F statistic follows a ࣲ^ଶ distribution. The package metap was

this calculation. Figure 6.4 shows the result, where five Type II

re identified when the replicate number was small for FKBP15

ype I errors presented when the replicate number was small for

1.