Biomedical Engineering Reference
In-Depth Information
Table 1
. Consistency of the expression levels for the genes selected
from the Whitehead Institute (WI) data in the Columbia University
data (CU) corresponding to Figure 2
No. of genes No. of genes
No. of genes consistent in inconsistent
Profile in WI data in CU data in CU data
p
-value
1 + 10
-20
DLBCL up/
129
115 (89%)
14 (11%)
FL down
FL up/
6 + 10
-11
81
69 (85%)
12 (15%)
DLBCL down
Overall
5 + 10
-30
210
184 (88%)
26 (12%)
profile
Three groups of genes are considered: those that are upregulated in diffuse large B-cell
lymphoma (DLBCL) over follicular lymphoma (FL), those with the opposite behavior,
and the union of the previous two sets. 88% of the genes selected from the WI data show
a consistent behavior in the CU data. The likelihood that this occurs by chance is ex-
tremely low, as shown in the
p
-value column.
domain, and because of the meaningful nature of the exercise. Testing in an in-
dependent data set the genes selected in our data is basically a way to do diag-
nostics. In other words, if we trust our methods of data mining, and the gene
expression array technology, then the genes found to express differentially in a
data set can be used as markers for diagnosis. If the chosen genes are truly in-
formative about the disease under study, one should in principle be able to de-
cide whether an array with a sample from a previously unseen subject corre-
sponds to an individual affected by the disease or to a healthy individual. In this
section we will present proof that this can be successfully done.
In (45), we used univariate and multivariate methods to discover genes that
differentiate between chronic lymphocytic leukemia (CLL) patients and normal
B cells. CLL is the most common leukemia in the United States and is a signifi-
cant cause of morbidity and mortality in the older adult population. The underly-
ing cause of CLL, however, remains unknown. In this regard, gene expression
profiling has been used successfully by a variety of investigators to discover
genetic differences between tumor cells and normal counterpart cells (45,52,53).
This information is proving to be very useful in understanding tumor cell biol-
ogy. In (45), Affymetrix Hu95A GeneChips were used to profile the gene ex-
pression from 38 CLL patients and from 10 healthy age-matched individuals to
identify key genetic differences between CLL and normal B cells. The univari-
ate methods selected 37 gene probes, whereas the multivariate method yielded
54 gene probes. Only 10 of the 81 total probes were identified in com-
