Biomedical Engineering Reference
In-Depth Information
between arrays E1 and E2 versus arrays E3 and E4. The pattern suggests
that the gene CreX 5 has dierential expression between the two dierent
experiment conditions (i.e., arrays E1 and E2 versus arrays E3 and E4).
1
5
14
1
16
6
11
12
7
2
9
10
19
4
8
18
1
13
17
3
1
14
1
16
11
5
6
14
1
16
12
7
2
5
6
11
9
10
2
7
19
12
8
18
9
10
4
3
19
17
18
4
8
13
20
3
13
17
20
20
6
8
10
12
14
6
8
10
12
14
6
8
10
12
14
E1
E1
E1
Figure 3. 1
Figure 3. 2
Figure 3. 3
1
1
1
14
1
16
14
11
5
11
6
15
16
6
14
15
16
5
2
7
12
12
2
7
11
6
9
10
10
2
7
12
8
19
1
19
8
18
9
10
4
3
4
3
19
17
17
18
8
4
13
13
3
13
17
20
20
20
6
8
10
12
14
6
8
10
12
14
6
8
10
12
14
E2
E2
E3
Figure 3. 4
Figure 3. 5
Figure 3. 6
Fig. 3. Verication of dierentially expressed genes. Figures 3.1-3.6 are the 6 pair-wise
comparisons between the four arrays, E1-E4. Figures 3.1 and 3.6 show smaller within
group variation. In contrast, large between group variation is graphically observed in
gures 3.2-3.5.
2.2. A Probe Rank Approach for Gene Selection
2.2.1. Rank Normalization for PM Intensity
Rank has been used as a normalization tool in microarray data analysis
61;62
,
but its use was limited to the gene level data. We extend its application to
the probe level data in oligonucleotide arrays. Rank avoids assumptions on
distribution of intensity. Rank also provides a better treatment for alleviat-
ing eects of extreme values. Our experiences from various microarray data
analyses have found that probe intensities in the oligonucleotide array data
often show a skewed distribution in extremely high intensities. In this case,
rank is more robust because the ranks of these extreme values are less in-
uential than their intensity levels. In addition, unlike other measurements
Search WWH ::
Custom Search