Learning Module Networks from Expression Profiles.
Eran Segal (1), Dana Pe'er (2), Aviv Regev (3), Daphne Koller
(1), Nir Friedman (2)
(1) Computer Science, Stanford university, Gates building 1A, Stanford,
CA 94305, USA;
(2) Hebrew University;
(3) Weizmann Institute
Functional modules are a key building block in the organization of a
cell. They consist of genes that are that are co-regulated by the same
'program' across multiple conditions. We propose a novel approach for
discovering modules and the control programs that govern them directly
from gene expression data. This approach is based on a new language of
module networks, an approach that partitions genes into modules: sets of
genes that act in a coordinated way, and whose behavior, as a group, can
be explained by common control rules. The control rules governing a
module encode the combinatorial and context-specific dependence of the
genes in the module on the expression level of some set of control genes
and on the experimental conditions. We develop a learning algorithm that
simultaneously learns assignments of genes to modules and the control
rules for each module. We apply this procedure to two yeast gene
expression datasets, and show that it recovers global modular
organization of genes, as well as detailed regulatory logic for these
modules. For example, one module consists of 22 genes, 21 of which are
part of the oxidative phosphorylation pathway, and which cover all six
components of the pathway. The primary regulator suggested by the
program for this module is HAP4, a known transcriptional regulator of
oxidative phosphorylation. Furthermore, motif analysis revealed a strong
presence in the genes' promoter region for a motif that is a known
binding site for HAP4.
Return to YGM 2002 Home at SGD