Biology Reference
In-Depth Information
is assigned to the votes, and then the sum of the votes {11
×
(
+
1)
+
6
×
(
−
1)
+
5, which means there are
more votes favoring the normal class, sample N28 is classified as “nor-
mal” and is given a class code of
6
×
0
=+
5} is calculated. Since the sum is
+
+
1. Otherwise, the sample should be
classified as “tumor” with a code of
1 if the sum of the votes is a nega-
tive value. The absolute value of the sum (ranging from 1 to 23) indicates
the predicting strength. When the sum equals zero, the sample under test
is unclassifiable.
For the sum of votes in the colon cancer data, one adds up the vote
values across all 23 gene subsets for every sample. If the sum of
the 23 prediction votes has a positive value (which indicates that there
are more gene subsets favoring the normal class than the tumor class), the
corresponding sample is classified as a normal sample with a code of 1. If
the sum of the 23 prediction votes has a negative value (which indicates
that there are more gene subsets favoring the tumor class than the normal
class), the corresponding sample is classified as a tumor sample with
a code of
−
1. Where the sum of the 23 prediction votes equals zero, the
corresponding sample is unclassified with a code of 0.
For the sum of votes in the leukemia data, one adds up the vote values
across all 23 gene subsets for every sample. If the sum of the 23 prediction
votes is positive (which indicates that there are more gene subsets favor-
ing the ALL class than the AML class), the corresponding sample is clas-
sified as an ALL sample with a code of 1. If the sum of the 23 prediction
votes is negative (which indicates that there are more gene subsets favor-
ing the AML class than the ALL class), the corresponding sample is
classified as an AML sample with a code of
−
1. Where the sum of the
23 prediction votes equals zero, the corresponding sample is unclassified
with a code of 0.
−
5.5. Discussion
Optimization algorithms are playing a significant role in the field of gene
selection and sample classification for microarray data. Many advanced
local and global optimization techniques, such as clustering and genetic
algorithms, have been successfully applied to gene subset selection for
classifying cancer tissue samples. Any optimization algorithm applied to
Search WWH ::
Custom Search