1. Bioinformatics Unit, ISTECH Inc.
A graph-theoretic modeling on GO space for biological interpretation of gene clusters
Bioinformatics Unit, ISTECH Inc.
Cancer Metastasis Research Center, Yonsei University College of Medicine
Sung Geun Lee,
Jung Uk Hur and
Yang Seok Kim
報告人:張家榮

2. Introduction Gene Ontology (GO) Clusters of microarray data
Controlled vocabulary of various genomic databases about diverse species
Clusters of microarray data
Each cluster has some genes
Extracts GO terms for a gene cluster
Each gene has several corresponding GO terms
Purpose
To discover the meaning of each cluster

4. Metric structure of GO tree

5. Lowest common ancestor
Given a non-empty subset U, v is a common ancestor of U if every node in U is on a subtree having v as the root and v0 is an LCA of U if v0 is greater than or equal to the level of w for any common ancestor w of U.

6. Lowest common ancestor

7. Principal distance Each level has its own weight W: IH -> R+
W(i) > W(i+1)
For example: W(k)=150-10(k-1)
Where w0 is LCA of v1 and v2

8. Principal distance 40 30 20 10

9. Multiset
Mathematically, the following three sets {1}, {1, 1}, {1, 1, 1} are equal in the set notation.
Yet, we want to take the number of occurrences of elements into account.
Such set is called as a multiset.

10. MaxPd and AverPd
given a multiset G ={v1, v2, , vn}

11. MaxPd and AverPd MaxPd AverPd
give the comprehensive biological meanings of a gene cluster
Not flexible but informs us of the existence of some functional outliers
AverPd
Signifies the most frequent GO codes
More than one

12. Algorithmic approach (1)
c[i,j] is j’st GO code of i’st gene
We consider ordered GO codes g[m] where 1≤m≤α
α is a constant related to the input data
α≤Ω , Ω is the total number of GO code

13. Algorithmic approach (2)
MaxPd is used to find LCA of C
Complexity : 3αn
min

14. MaxPd 40 30 20 10 40 30 40 40 20

15. Algorithmic approach (3)
AverPd is used to find an optimal GO code g[m0] such that the average distance between g[m0] and each gene in C is smaller than that of any g[m]
Complexity : 3αn

16. AverPd

17. Discussion
Other algorithms consider GO term frequencies or compare specific GO term-related gene groups
In our modeling, the topological property of GO hierarchy is used.

18. Utility
Biological assessment of the clustering results of DNA microarray data
Coupled with any clustering technique to predict the functional category of the unknown genes
Not only DNA microarray data, but also any kinds of group analysis with any ontology having an identical structure with GO

19. Another approach The length of GO code is about logα
Take one number of GO code each time
pseudo-code:
for 1≤k≤ logα
cluster C by the kth number
break if no cluster above n
delete clusters under n
end

20. MaxPd

21. Complexity Worst case : O( nlogα) Best case : O (n) Alternate
Change the break time to cluster it in detail

22. disadvantage
Cannot obtain information not contained in GO such as disease-related genes
GO terms on the same level have different level information
GO hierarchy is dynamic and flexible