Biology Reference
In-Depth Information
gives the contours for the Fisher combined
p
-statistic, while Fig. 1(c)
shows contours for the Venn diagram rule. Figure 1(d) shows a closeup,
providing a sense of how all three rules are related (
0.001 only). The
Venn diagram rule gives up much power by stringent within-study
requirements. While the Fisher rule picks up genes that provide moder-
ate to strong evidence in both studies, it also calls significant genes with
very strong positive evidence in one study yet moderate to strong nega-
tive evidence in the other (points in quadrants II and IV). This is essen-
tially because these points are closer to quadrant I (the alternative
hypothesis) than to the origin (the null). The test based on
-
does not
suffer so much from this drawback, and picks up some power in the range
of roughly equal moderate evidence in both studies.
Table 3 gives the
z
-scores from individual studies for genes with
-
α =
>
8
and
-
7. Genes not measured in an individual study have blank
entries. The table also shows the number of overall top and bottom genes
occurring in the top and bottom 100 genes in each individual study. The
utility of combining results is clearly demonstrated here.
<−
7.2. Example II:Breast Cancer Gene Signatures
In the previous example, we focused on combining information on the
effect on survival of expression of single genes. Here, we show how meta-
analysis can also be used to integrate patterns of gene expression, or gene
“signatures”. Signatures can be useful for tumor subtyping, diagnosis, or
prognosis.
Disparate breast cancer gene expression signatures have been pro-
posed, with little agreement in the constituent genes.
30,32,33,37,40,44-46,52
We show how our meta-analytical approach can be used to uncover
relationships that are consistent in a large collection of public datasets,
and are thus unlikely to be artifacts of specific cohorts or microarray
platforms.
We introduce the concept of “coexpression modules”, comprehen-
sive lists of genes with highly correlated expression, which we use to ana-
lyze gene expression signatures.
