Transcript network_AML_2016_11_4x

Network analysis for AML data
Dr. Habil Zare, PhD
Oncinfo Lab
Texas State University
14 Jan 2015
Bayesian networks are useful in
modeling genes interactions
2
CCLE provides expression data useful
for learning the network
3
First application
Biological interpretation
Gene hubs, causal relationships,
interaction between pathways, ….
4
Data cleaning
FKPM values less than 1
Are considered noise and
technical artifact.
5
Differential expression
We used limma and ranked the transcripts based
on their differential expression in AML vs. MDS.
We used the top third for further analysis that
included 913 non-coding RNAs.
Length of non-coding RNAs
6
Gene modules
Size of module identified by WGCNA.
44 AML cases
24 MDS cases
The difference is the number of size of modules is
most likely due to different sample size in each group
7
Gene modules
30 random
30 random
AML cases
MDS cases
MDS is expected to have more modules because of
higher heterogeneity.
8
Gene hubs in modules
Hubs are the transcripts that have the highest
correlation (interaction) with other transcripts in
a module.
9
Gene hubs in modules
We looked at the hubs in the modules and found that:
A)
PBX3, HOXA9, HOXA10, and MEIS1 are among the top 10 hubs in
AML-36 module. They are know to be associated with AML.
PBX3 is a cofactor for HOXA9 and both of them are targets for AML
therapy. The cluster of HOXA genes are known to be very well
associated with leukemia; "Artificial overexpression of HOXA7, HOXA9,
or HOXA10 in combination with MEIS1 caused leukemia in animal
models.” (Bach et al. 2010)
10
Gene hubs in modules
We looked at the hubs in the modules and found that:
B) 9 isotopes of HLA-DPA1 are hubs of AML-20 reported to be associated
with AML in 10 studies according to Miler 2010. HLA-DPA1 is also a hub
in MDS-35 module.
11
Gene hubs in modules
C) The other hub of AML-20 module is HLA-DMA. Can HLA-DMA be
interesting for us too? They both have a high correlation of 0.94 over all
AML and MDS cases.
12
Gene hubs in modules
HLA-DPA1 and HLA-DMA are under-expressed in AML.
HLA-DMA was reported in 5 studies in the review paper but is its role in
AML understood as much as the other gene in this module?
13
Genes in module 36 correlate
differently in AML vs. MDS
AML
MDS
14
Distribution of 4000 known,
AML-related genes in the modules
Other than module 36, module 46 that is also enriched in interesting
genes.
15
Genes in module 46 correlate
differently in AML vs. MDS too
AML
MDS
16
One striking difference
2 HOXB cluster antisense RNA, NR_033202 and NR_033201, have
almost no correlation in MDS.
AML
MDS
17
NR_033202 and NR_033201
~0.5 Kbs long. Both are expressed in more than 1/3 of AML
cases and rarely in MDS case. Are they associated with an
AML subtype?
18
NR_033202 and NR_033201
They are associated with therapy-related AML or MDS.
NR_033201:
•
40% of tAML cases have expression > 1 (above noise level) while
subtypes AML-MEGA, AML-MPN, AML-NK, MDS have 0 expression.
•
Expressed in AML-M6:1 AML-MDS :1 tAML:12 tMDS : 2
NR_033202:
•
AML-M6:1 AML-MDS:3
tAML:11 tMDS:3
19
Next steps
1. Controlling for confounding variables such as age and sex
2. Study other gene hubs
3. Learning a Bayesian Network for each module
and studying the most influential genes
(recognized as N-hob-down-stream nodes)
4. Introducing a new binary variable to the network that indicates
whether the sample is AML and MDS. The genes connected to this
new random variable are useful in explaining the biological
difference between AML and MDS.
20