Transcript No Slide Title

Joint analysis of three seemingly
independent microarray experiments
via multivariate mixed-model equations
with null residual covariance structure
Antonio Reverter1, Yong-Hong Wang1, Keren A Byrne, Siok Hwee Tan1,
Gregory S Harper1, Heather L. Bruce2, and Sigrid A Lehnert1
The Cooperative Research Centre for Cattle and Beef Quality
1CSIRO
Livestock Industries, Queensland Bioscience Precinct
306 Carmody Rd, St Lucia, QLD 4067, Australia
2Food Science Australia, Tingalpa DC, QLD 4173, Australia
applied across experiments yielding 1,424,322 intensity records analysed by
fitting a tetra-variate model with 263,938 equations and 18 (co)variance
components. The latter were estimated by restricted maximum likelihood.
Correlation estimates for gene expressions ranged from 0.477  0.004 (for EXP1
and EXP3) to 0.896  0.002 (for EXP1 and EXP2). These moderate to strong
estimates indicate the power to be gained with the joint analysis. The expected
percentage of differentially expressed genes, as measured from the gene by
treatment interaction, was 1.2% for diet in EXP1, 1.3% for breed in EXP2, and
17.1% and 5.1% for age effect and adipogenic treatment, respectively in EXP3.
ABSTRACT: A bovine microarray of muscle and fat cDNAs was used in three
gene expression experiments (EXP1, EXP2, and EXP3). EXP1 (14 slides)
contrasts the gene expression profiles of muscle in Brahman steers fed varying
quality diets. EXP2 (12 slides) compares the expression profile in muscle tissue
between two breeds. EXP3 (22 slides) studies the mechanisms underlying
adipogenesis in vitro. This study undertook to jointly analyse these three
experiments using multivariate mixed-models with the relaxed assumption of a
non-zero correlation amongst gene expressions across experiments while
imposing a null residual covariance structure. Equivalent editing criteria were
EXP1: Rockhampton Diets (LD Muscle)
MEDIUM
(4  Animals)
Same Microarray Slide
Across the 3 Experiments
(48 slides)
LOW
(3  Anim, 1 Rep)
(pooled 3 Anim)
HIGH
(pooled 2 Anim)
MEDIUM
(Pool & Ampl)
CORRELATION:
EXP2: Marbling Breeds (LD Muscle)
TIME 2
There is gain to be made
from a joint analysis
MODEL: y  X  Zg  Mt  e
LOW
(Pool & Ampl)
TIME 1
EXP1 & EXP2 = 0.84 (7,515)
EXP1 & EXP3 = 0.56 (8,717)
EXP2 & EXP3 = 0.51 (7,316)
 X T R 1 X
 T 1
Z R X
 M T R 1 X

X T R 1 Z
Z T R 1 Z  G 1  I
T


T
1
Z R M

M T R 1 M  T 1  I 
X T R 1 M
EXP1
EXP2
Tot+AmpRNA
AmpRNA
1,344
9,245
27,678
405,820
1,152
9,245
45,096
329,247
672
9,245
45,474
219,956
1,440
9,245
104,102
469,299
9.3
0.1
1.4
9.7
0.1
1.5
3.8
1.5
3.3
8.8
0.6
2.0
Differentially Expressed
% Total Variation
1.2
Actual N (95% CI) 403
1.3
354
17.1
n/a
5.1
546
1
M R Z
E  0  i  j
TIME 3
EXP3
TotRNA
AmpRNA
i, j
BREED 1
BREED 1
BREED 1
BREED 2
BREED 2
BREED 2
Log(Intensity) = Systematic +
Gene
+
Gene*Trt +
Error
4 Variables
1,424,322 Records
263,938 Equations
18 (co)variances
EXP3: Adipogenesis (in vitro)
REML
V(G)
V(G*T)
V(E)
T1
T2
T3
T4
T5
RESULTS:
Jointly Differentially Expressed
T6
TREAT 1
TREAT 2
IMPLICATIONS
Go beyond standard applications:
Pleiotropic
Patterns of Co-Regulation
Evolution
Plasticity
Diet Restriction Up
Down
Breed 1
Breed 2
Treat 1
Treat 2
Up
Down
84
0
319
Breed 1
Breed 2
Treat 1
Treat 2
2
20
1
5
2
9
1
12
121
0
233
2
15
2
2
403
0
143