Causal Inference and the Construction of Predictive Network Models in Biology - Systems Biology Concepts and Insights - page 485

A) B)

AFG1 AFG1 AFG1

dihydroorotic_acid dihydroorotic_acid dihydroorotic_acid

dihydroorotic_acid dihydroorotic_acid dihydroorotic_acid

YEL016C YEL016C YEL016C

orotic_acid orotic_acid orotic_acid

orotic_acid orotic_acid orotic_acid

AFG1 AFG1 AFG1

YEL041W YEL041W YEL041W

YHC1 YHC1 YHC1 YGL 185C YGL 185C YGL 185C

URA1 URA1 URA1

MCM1 MCM1 MCM1

URA1 URA1 URA1

YHC1 YHC1 YHC1

MCM1 MCM1 MCM1

MDH2 MDH2 MDH2

MRE11 MRE11 MRE11

URA4 URA4 URA4 URA2 URA2 URA2

URA3 URA3 URA3

URA3 URA3 URA3

RIB4 RIB4 RIB4

URA2 URA2 URA2

YEL016C YEL016C YEL016C

RPN13 RPN13 RPN13

YHM2 YHM2 YHM2

AQR1 AQR1 AQR1

MOH2
MOH2 MOH2

YHM2 YHM2 YHM2

URA4 URA4 URA4

RIB4 RIB4 RIB4

RPN13 RPN13 RPN13 YEL041W YEL041W YEL041W

AQR1 AQR1 AQR1

C) D)

16

14

YEL041W YEL041W YEL041W

dihydroorotic_acid dihydroorotic_acid dihydroorotic_acid

12

MDH2 MDH2 MDH2

AFG1 AFG1 AFG1

orotic_acid orotic_acid orotic_acid

10

URA2 URA2 URA2

URA1 URA1 URA1

8

6

URA4 URA4 URA4

YHC1 YHC1 YHC1

RPN13 RPN13 RPN13

URA3 URA3 URA3

4

2

YHM2 YHM2 YHM2

RIB4 RIB4 RIB4

MCM1 MCM1 MCM1

0

1 3 456 789

chromosome

10 11 12 13 14 15 16

AQR1 AQR1 AQR1

FIGURE 26.4 Example yeast network. (A) Subnetwork identified in a previously constructed whole genome yeast network in which URA3 was
predicted as the causal regulator for genes and metabolites linked to a genetic locus on chromosome 5 coincident with the physical location of URA3. Red
nodes are genes or metabolites whose variations are linked to the chromosome V locus. Hexagonal nodes represent metabolites, circular nodes represent
genes, and diamond-shaped nodes represent genes with cis eQTLs. (B) Trait values of nodes compared with genotype data for the URA3 subnetwork.
eQTLs and metQTLs are prominently featured as residing in the chromsome V URA3 locus. (C) Bayesian network reconstructed using only trait data.
(D) Bayesian network reconstructed using trait data and priors derived from other types of data.

Step 1: Identification of the URA3-Centered de
novo Biosynthesis of Pyrimidine Ribonucleo-
tides Subnetwork
There are 18 nodes in the subnetwork shown in Figure 26.4 a.
These nodes are highly correlated with one another, with
68% of all pairwise relationships significant at the 0.01
significance level. The continuous gene expression data for
these 18 genes can then be discretized into three states
representing downregulated, no change, and upregulated,
and the mutual information of all pairs of nodes are then
calculated. In this case, 54% of all pairs are significant at
p

used to assess the significance of the mutual information of
the observed data). All 18 of the trait values corresponding to
these nodes are significantly associated with the genotypes at
the URA3 locus ( Figure 26.4 c).

Step 2: Reconstructing Networks using only
Expression and Metabolite Traits (Excluding
DNA Variation Data)
The process of reconstructing networks using only trait data
is straightforward. The trait data are input into a standard
Bayesian network reconstruction program in which 1000
network structures are generated from a Monte Carlo
Markov Chain (MCMC) process using different random

0.01 (the mutual information of the permutated data is
calculated and fit into a normal distribution, which is then

<