Biology Reference
In-Depth Information
The correlated expression measures in a module provide redun-
dant information about the module's overall expression in a tumor.
They can be summarized into a single number by averaging. We
called the resulting value a “module score”. In addition to examining
genes in the modules, we can study properties of the module scores,
for example, their effectiveness in tumor subtyping or prognostic
value. We have found that this type of examination can produce
not only effective tumor classifications or predictions, but also new
biological insights.
Focusing for now on AURKA, we look for an association with
survival. For each gene, we combine the
z
-scores from Cox regression
[Eq. (2)] according to Eq. (1), and consider these as a function of the
combined
z
-score for the AURKA regression coefficient [Eq. (3)].
Figure 3(a) shows single-gene
z
-scores for the AURKA coefficient
within the NKI and EMC signature datasets. The genes corresponding
to each signature are highlighted in the plot, showing that the genes
comprising the signature are largely different and also span the range of
Z
-AURKA.
Fig. 3.
Gene signature prognostic performance for NKI (o), EMC (x), and all
datasets combined (
). (a) Within-dataset AURKA
z
-scores for NKI and EMC;
(b) overall
-
for survival vs.
-
-AURKA.
•
