=Paper=
{{Paper
|id=Vol-1747/IP31_ICBO2016
|storemode=property
|title=Planteome Gene Annotation Enrichment Analysis
|pdfUrl=https://ceur-ws.org/Vol-1747/IP31_ICBO2016.pdf
|volume=Vol-1747
|authors=Botong Qu,Jaden Diefenbaugh,Eugene Zhang,Justin Elser,Pankaj Jaiswal,Seth Carbon,Christopher Mungall
|dblpUrl=https://dblp.org/rec/conf/icbo/QuDZEJCM16
}}
==Planteome Gene Annotation Enrichment Analysis ==
https://ceur-ws.org/Vol-1747/IP31_ICBO2016.pdf
Planteome Gene Annotation Enrichment Analysis
Botong Qu Justin Elser Seth Carbon
Jaden Diefenbaugh Pankaj Jaiswal Chris Mungall
Eugene Zhang Department of Botany Lawrence Berkeley
School of Electrical and and Plant Pathology National Laboratory
Computer Engineering Oregon State University Berkeley, CA 94720
Oregon State University Corvallis, Oregon 97331
Corvallis, Oregon 97331
Abstract—Annotation enrichment analysis of a gene list helps This 2 by 2 table is the contingency table used to calculate the
biologists to identify the potential biological functions associated p-values for each term. If the p-value is bigger than the user
with it. With the extensions of plant ontology categories, the chosen cut-off value (0.01 or 0.05), the term is not enriched
discovery of significant ontology terms associated with a gene list
becomes more and more informative. We introduce a tool to help by the gene list.
biologists to find out these terms based on the expanding ontology
database of the Planteome project. In addition, we propose some TABLE I
new visualization schemes to help users construct a meaningful T HE CONTINGENCY TABLE FOR A ONTOLOGY TERM
interpretation of the results guided by the ontology tree.
Input Genes Not Input Genes Sum (Ref)
keywords—Ontology, Plants, Term enrichment analysis, Visu-
Annotated m k-m k
alization Not annotated n-m (N-n)-(k-m) N-k
Sum n N-n N
I. I NTRODUCTION
Gene annotations are analyzed and explored by gene cura- With the number of genes annotated to the term inside the
tors from all over the world. Finding and visualizing the useful gene list (m), the total number of genes annotated to the term
information from the annotations has been a hot topic for in the whole database (k), the number of input genes (n) and
decades. The Common Reference Ontologies and Applications total number of genes in the database (N ), Fisher’s exact test
for Plant Biology benefits biologists to be able to discover is defined as equations 1 and 2. The H(m, k, n, N ) represents
enriched biological ontology terms among all provided on- the hypergeometric distribution.
tologies (Gene Ontology, Plant Ontology, Trait Ontology, k
N −k
Environment Ontology, etc.). To assist this analysis process, m × n−m
H(m, k, n, N ) = N
(1)
we provide a gene annotation enrichment analysis tool which n
uses Fisher’s exact and chi-squared methods to statistically k
X
analyze all annotation data. Then, we visualize the results two p − value = H(i, k, n, N ) (2)
ways: 1) Highlighting the enriched terms among all ontology i=m
terms in the database to emphasize relative positions of the Based on the contingency table, we calculate the expected
enriched terms. 2) Considering the cut-off p-value as a basis value of the cell that represents the number of genes annotated
of an uncertainty factor when visualizing the tree structure in k
to the term and inside the input list by n× N , then we construct
order to conveniently focus on the interesting terms. an expected contingency table by fixing the margin values k, n,
and N and using the calculated expected value to calculate all
II. A NALYSIS M ODEL AND M ETHODS other three cells. Then we calculate the χ2 value with equation
In our tool, we provide two common analysis methods to 3, and then transfer it to p-value for 1 degree of freedom (a 2
find the enriched terms: the Fisher’s exact test and the chi- by 2 table always has a freedom of 1).
squared test [1]. To apply these statistical analysis methods, X (expcted − observed)2
the formulation of a contingency table is necessary. In our χ2 = (3)
expcted
system, we create the contingency table (table I) similar to all cells
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
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. Fig. 2. A hair-ball style visualization of mega data
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
Fig. 1. Analysis result of a set of genes information.
III. A NALYSIS V ISUALIZATION
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)
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
(b)
Biological ontology terms are always organized in a hierar-
chial structure, i.e. each ontology term inherits the properties Fig. 3. a) visualization from AgriGO (b) uncertainty visualization with re-
of their parents and differs with its siblings in some functional- arranging the layout
ities. Since each ontology term can have multiple parents and
siblings, the research of the enriched ontology branch of a set
R EFERENCES
of genes facilitates biologists to explore the potential functions
associated to the genes and can be applied to find featuring [1] D. W. Huang, B. T. Sherman, and R. A. Lempicki, “Bioinformatics
enrichment tools: paths toward the comprehensive functional analysis of
genes in it. To visualize the enriched branch, we apply a hair- large gene lists,” Nucleic acids research, vol. 37, no. 1, pp. 1–13, 2009.
ball style visualization (similar to Fig. 2) to all the ontology [2] I. Rivals, L. Personnaz, L. Taing, and M.-C. Potier, “Enrichment or deple-
terms included in the database and highlight the ones that are tion of a go category within a class of genes: which test?” Bioinformatics,
vol. 23, no. 4, pp. 401–407, 2007.
significant to our input genes. [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,”
B. Uncertainty Visualization PloS one, vol. 7, no. 10, p. e46128, 2012.
The hierarchical visualization of the analysis results (e.g. [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,
Gene Ontology terms) is a common method to facilitate users 2010.
to explore the biological meanings behind the gene lists [4]. [5] E. Eden, R. Navon, I. Steinfeld, D. Lipson, and Z. Yakhini, “Gorilla: a
The tree structure graphs (as Fig. 3a shows) describe the tool for discovery and visualization of enriched go terms in ranked gene
lists,” BMC bioinformatics, vol. 10, no. 1, p. 48, 2009.
hierarchical structured ontology terms pretty well and have [6] S. Maere, K. Heymans, and M. Kuiper, “Bingo: a cytoscape plugin
a common use in analysis tools ( [4], [5], [6]). However, to assess overrepresentation of gene ontology categories in biological
there are some shortages in these visualization results. For networks,” Bioinformatics, vol. 21, no. 16, pp. 3448–3449, 2005.