Planteome Gene Annotation Enrichment Analysis

Abstract

-Annotation enrichment analysis of a gene list helps biologists to identify the potential biological functions associated with it. With the extensions of plant ontology categories, the discovery of significant ontology terms associated with a gene list becomes more and more informative. We introduce a tool to help biologists to find out these terms based on the expanding ontology database of the Planteome project. In addition, we propose some new visualization schemes to help users construct a meaningful interpretation of the results guided by the ontology tree.

-

In our tool, we provide two common analysis methods to find the enriched terms: the Fisher’s exact test and the chitsshyqesutaefrmoerd,mwtuelesattci[or1en]a.toeTfotahaepcopcnloytnitntihgneegsneecnycsytatattibasblteilceaisl(taannbealceleysIss)iasrsymi.meIitnlhaorodutsor, 2 = allXcells (expcteedxpcotbesderved)2 (3) ones used in [ 2 ] and [ 3 ]. For one specific ontology term and Each of these two methods has its own strengths and n genes, all genes in the database (N ) are classified into four weaknesses. The Fisher’s exact test can be applied when the categories: the genes annotated to the term and in the input input genes number is small and provides an exact calculation gene list (m), the genes not annotated to the term and in the of the significance of the null hypothesis. But when the sample input gene list (n m), the genes annotated to the term and is large or the data is well balanced, the Fisher’s exact test not in the input gene list (k m), the genes not annotated becomes computationally costly for the factorial calculation to the term and not in the input gene list (N n k + m). involved. On the other side, the chi-squared test can be applied

I. INTRODUCTION

Gene annotations are analyzed and explored by gene curators from all over the world. Finding and visualizing the useful information from the annotations has been a hot topic for decades. The Common Reference Ontologies and Applications for Plant Biology benefits biologists to be able to discover enriched biological ontology terms among all provided ontologies (Gene Ontology, Plant Ontology, Trait Ontology, Environment Ontology, etc.). To assist this analysis process, we provide a gene annotation enrichment analysis tool which uses Fisher’s exact and chi-squared methods to statistically analyze all annotation data. Then, we visualize the results two ways: 1) Highlighting the enriched terms among all ontology terms in the database to emphasize relative positions of the enriched terms. 2) Considering the cut-off p-value as a basis of an uncertainty factor when visualizing the tree structure in order to conveniently focus on the interesting terms.

II. ANALYSIS MODEL AND METHODS

This 2 by 2 table is the contingency table used to calculate the p-values for each term. If the p-value is bigger than the user chosen cut-off value (0.01 or 0.05), the term is not enriched by the gene list.

Input Genes

m n-m n

Not Input Genes

k-m (N-n)-(k-m)

N-n

Sum (Ref) k N-k N

With the number of genes annotated to the term inside the gene list (m), the total number of genes annotated to the term in the whole database (k), the number of input genes (n) and total number of genes in the database (N ), Fisher’s exact test is defined as equations 1 and 2. The H(m; k; n; N ) represents the hypergeometric distribution.

H(m; k; n; N ) = p

k value = X H(i; k; n; N ) k m

N n

N k n m (1) (2) i=m

Based on the contingency table, we calculate the expected value of the cell that represents the number of genes annotated to the term and inside the input list by n Nk , then we construct an expected contingency table by fixing the margin values k, n, and N and using the calculated expected value to calculate all other three cells. Then we calculate the 2 value with equation 3, and then transfer it to p-value for 1 degree of freedom (a 2 by 2 table always has a freedom of 1). to large data samples but can only give an approximation of the significance. Both methods could be used to reject the null hypothesis that the data are independent, i.e. the input genes don’t enrich the ontology term.

After inputting an interesting gene list, the server will query graphically among all the annotation data, i.e. transfer all annotations of an ontology term to its parents to make sure the indirect annotations are involved in the analysis. The final analysis is as shows in Fig. 1.

III. ANALYSIS VISUALIZATION

Besides the detail information of enriched ontology terms, there are two other kinds of information to be explored. First, the relationship among enriched terms and their corresponding significant levels, the significant levels are not only limited to the p-values, but also the number of input genes annotated to a particular ontology term. Second, the relationship between enriched terms and the whole reference data. We want to apply two visualization methods to help users efficiently perceive these information.

A. Enriched Ontology Branch Visualization

Biological ontology terms are always organized in a hierarchial structure, i.e. each ontology term inherits the properties of their parents and differs with its siblings in some functionalities. Since each ontology term can have multiple parents and siblings, the research of the enriched ontology branch of a set of genes facilitates biologists to explore the potential functions associated to the genes and can be applied to find featuring genes in it. To visualize the enriched branch, we apply a hairball style visualization (similar to Fig. 2) to all the ontology terms included in the database and highlight the ones that are significant to our input genes.

B. Uncertainty Visualization

The hierarchical visualization of the analysis results (e.g. Gene Ontology terms) is a common method to facilitate users to explore the biological meanings behind the gene lists [ 4 ]. The tree structure graphs (as Fig. 3a shows) describe the hierarchical structured ontology terms pretty well and have a common use in analysis tools ( [ 4 ], [ 5 ], [ 6 ]). However, there are some shortages in these visualization results. For example, in Fig. 3(a), the relative low significant terms (less red ones) could be distractive if the users only want to focus on the most significant terms. Also, the fixed cut off p-values make the visualization results not flexible enough. Therefore, it would be useful if we consider the cut-off p-value as a uncertainty factor and graph it. In this way, users are able to set an interesting significance value range, then the visualization results will re-arrange the focused terms to the center (as shown in Fig. 3b) to help biologists easily study them. The structure relationship between them and the significant levels calculated are always preserved to provide correct hierarchical information.

References

[1] D. W. Huang , B. T. Sherman , and R. A. Lempicki , “ Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists , ” Nucleic acids research , vol. 37 , no. 1 , pp. 1 - 13 , 2009 .

[2] I. Rivals , L. Personnaz , L. Taing , and M.-C. Potier , “ Enrichment or depletion of a go category within a class of genes: which test?” Bioinformatics , vol. 23 , no. 4 , pp. 401 - 407 , 2007 .

[3] G. Mi , Y. Di , S. Emerson , J. S. Cumbie , and J. H. Chang , “ Length bias correction in gene ontology enrichment analysis using logistic regression,” PloS one , vol. 7 , no. 10 , p. e46128 , 2012 .

[4] Z. Du , X. Zhou , Y. Ling , Z. Zhang , and Z. Su , “ agrigo: a go analysis toolkit for the agricultural community , ” Nucleic acids research , p. gkq310 , 2010 .

[5] E. Eden , R. Navon , I. Steinfeld , D. Lipson , and Z. Yakhini , “ Gorilla: a tool for discovery and visualization of enriched go terms in ranked gene lists,” BMC bioinformatics , vol. 10 , no. 1 , p. 48 , 2009 .

[6] S. Maere , K. Heymans , and M. Kuiper , “ Bingo: a cytoscape plugin to assess overrepresentation of gene ontology categories in biological networks , ” Bioinformatics , vol. 21 , no. 16 , pp. 3448 - 3449 , 2005 .