1.  Identification of optimal parameter ranges in building and assessing correlation networks built from gene expression.
Qianran Li
Supervisor：Dr. Kathryn Cooper
Student Research and Creative Activity Fair
March 2nd, 2018

2. Background: Gene Expression
DNA is regulated by transcription factors
Transcription Factor
(Regulator)
DNA

3. Background: Gene Expression
Transcription Factor
(Regulator)
The Transcription Factor tells the genes when to make a protein
Genes

4. Background: Gene Expression
This process is called gene expression
Transcription Factor
(Regulator)
Proteins

5. Problem Volumes of gene expression data available
Goal of gene expression: Identify genes that cause disesase, aging….
Some methods are available (GSEA)
But, sometimes these methods return no results or too many to study
Hard to identify the functional differences of proteins among tissues

6. Background: Correlation Network
A correlation network is is built from gene expression data, and is used for examining gene co-expression
Image source: Newman, MEJ. (2002). Assortative Mixing in Networks. Phys. Rev. Let., 89(20):
Dempsey, K., Bonasera, S., Bastola, D., & Ali, H. (2011, January). A novel correlation networks approach for the identification of gene targets. In System Sciences (HICSS), th Hawaii International Conference on (pp. 1-8). IEEE.

7. Motivation Correlation networks have structural characteristics
Assortativity
Gene connectedness
Clustering coefficient
Does a group form a cluster of edges?
Image source: Newman, MEJ. (2002). Assortative Mixing in Networks. Phys. Rev. Let., 89(20):

8. Research Methodology
Get data from Gene Expression Omnibus (GEO).
Build the network based on their Pearson correlation coefficient
Develop structural analysis pipeline by using igraph package in R
Identify possible structural measurement ranges

9. Data Collection Gene expression networks built from Mus musculus
5 experiments (experiment = series)
35 networks
Series
Networks
Tissues
Sample per network
GSE38531 7
blood 5
GSE41127 8
T-cell
GSE38754
Heart, lung, kidney
GSE24637
liver, pancreas, muscle, Adipose
GSE27567 4
breast tumor 10

10. Network Creation Pearson Correlation coefficient
Filtered to only the highest of correlations
From |0.7 to 1.0|
Only correlations passing P-value < (Student’s T-test)
Dempsey K, Thapa I, Bastola D, et al. Identifying modular function via edge annotation in gene correlation networks using Gene Ontology search[C]//Bioinformatics and Biomedicine Workshops (BIBMW), 2011 IEEE International Conference on. IEEE, 2011:

11. Results: Assortativity Coefficient
Assortativity coefficient the Assortativity coefficient means the degree of co-expression between the hub genes in this network
Hub
Hub
Hub
Hub
Assortative
Disassortative

12. GSE38531
GSE41127
GSE38754
GSE24637
GSE27567

13. Results: Clustering coefficient
A Cluster is a group of nodes that are tightly interconnected.
The Clustering coefficient is the degree to nodes which tend to be a cluster.

14. GSE38531
GSE41127
GSE38754
GSE24637
GSE27567

15. Results: Range Identification
All
Mean
Median
Min
Max
# Networks
Nodes
37,468.6
45,096
37,299
45,101 35
Edges
6,101,447
2,765,449
257,877
74,542,976
Assortativity
0.58
0.59
-0.36
0.95
Clustering Coefficient
0.56
0.62
0.30
0.79

16. Results: Range Identification

17. Conclusion Built 35 networks from 5 experiments for “meta” analysis
Clustering coefficient varies across experiments
Significant variance in between-series clustering.
No other measures vary consistently across experiments
Clustering coefficient has significant variance between-series
Next steps
Add other parameter analysis such as degree distribution and hub node analysis.
Control number of samples per networks.
Add more data sets
We need further testing, but if we result is correct, means this model is ready for model gene expression data. which means the cell is not randomly pick the functions.

18. Acknowledgement Kate Cooper GEO database Igraph package in R
Bioinformatics lab members

19. Questions?