The biological networks controlling plant signal transduction, metabolism and gene regulation are composed of not only genes, RNA, protein and compounds but also the complicated interactions among them. Yet, even in the most thoroughly studied model plant Arabidopsis thaliana, the knowledge regarding these interactions are scattered throughout literatures and various public databases. Thus, new scientific discovery by exploring these complex and heterogeneous data remains a challenge task for biologists. We developed a graph-search empowered platform named HRGRN to search known and, more importantly, discover the novel relationships among genes in Arabidopsis biological networks. The HRGRN includes over 51,000 “nodes” that represent very large sets of genes, proteins, small RNAs, and compounds and approximately 150,000 “edges” that are classified into nine types of interactions (interactions between proteins, compounds and proteins, transcription factors (TFs) and their downstream target genes, small RNAs and their target genes, kinases and downstream target genes, transporters and substrates, substrate/product compounds and enzymes, as well as gene pairs with similar expression patterns to provide deep insight into gene-gene relationships) to comprehensively model and represent the complex interactions between nodes. . The HRGRN allows users to discover novel interactions between genes and/or pathways, and build sub-networks from user-specified seed nodes by searching the comprehensive collections of interactions stored in its back-end graph databases using graph traversal algorithms. The HRGRN database is freely available at http://plantgrn.noble.org/hrgrn/. Currently, we are collaborating the Araport team to develop REST-like web services and provide the HRGRN’s graph search functions to Araport system.

1. 1
HRGRN: enabling graph search and integrative analysis of
Arabidopsis signaling transduction, metabolism and gene
regulation networks
Speaker: Xinbin Dai
The Samuel Roberts Noble Foundation
Plant & Animal genome Conference (PAG), 2016
• Nodes: Genes, proteins, non-coding RNAs and small compounds
• Interactions among these elements
Complex biological networks
We need full knowledges of related biological interactions
to understand gene function
• The interaction data is heterogeneous
– Protein-protein interaction, TF-target, miRNA-target, enzyme-
compound, and transporter-target
– Generated from different experiment methods
• and is scattered
– Low-throughput experiment data. literature
– High-throughput experiment data. Microarray, RNA-seq, ChIP-
chip
– Predicted interactions using bioinformatics software. Gene co-
expression analysis and promoter motif analysis
Challenging of utilizing known biological interaction
data
Example: Is there any
relationship between gene
A and gene D?
Edge 1: Protein-protein
interaction (BioGrid/AtPIN
database)
Edge 2: co-expressed gene pair
(predicted)
Edge 3: TF-target gene
regulation (literature)
A
D
1
2
3
Difficulty in utilizing the heterogeneous and scattered biological
interaction data for gene function annotation
C
B
Biologists have to find the right database, literatures and also have to
analyze gene expression profiling data to retrieve the information.
• Node: gene, protein, small RNA, and compound
• Seven type of edges: Heterogeneous interaction data are organized by
biological definition:
– Protein-protein interaction: BioGrid and AtPIN database
– Compound-protein interaction: curation from literature
– Transcription factor-target: curation from literature
– Small RNA-target: prediction from psRNATarget and curation
– Transporter-substrate: transporter prediction and TCDB database
– Enzyme-compound: KEGG database
– Co-expressed gene pair: data analysis from transcriptomic data
• http://plantgrn.noble.org/hrgrn/, Plant Cell & Physiology 2015
HRGRN utilizes graph model to integrate these heterogeneous
and scatted interaction data
• Neo4j database: 3-4 orders of magnitude faster than
SQL database for graph path search between nodes
– Host node and interaction data
– Provide framework of graph search algorithm for
HRGRN in Java
• Cytoscape.js: a HTML5 JavaScript library on front
end, which display node/edge in browser.
• Customized code in Groovy/JavaScript (front-
end) and Java (graph search at the back-end)
Technical Implementation of HRGRN

2. 2
Case #1: HRGRN provides all associated nodes and interactions
of a specific gene in a graph of the neighborhood
• Individual gene-centered sub-network
• Graph traversal
algorithms in
Neo4j database
- Breadth-first
search
• Customized Java
code for our
defined edge
types and other
properties during
search
Predicted
relationship
Validated
relationship
Positive
Interaction
HRGRN use line color, line shape and arrow
shape to represent interaction properties
Negative
Interaction
Graph search and visualization are user-customizable
• Highlight node by keyword
• Change color for highlighting
• Toggle predicted/validated edge
display
• Change graph layout
• Export graph figure
Change graph path search option:
• Type of interaction by biological
definition
• Validate or predicted interaction
• Quantified similarity of co-
expression pattern between
genes
Case #2: HRGRN can discover “unknown” relationships
between genes
• The relationships between
ATCUL3 and IAA28 will be
skipped during a traditional
SQL database query
• Shortest path search
algorithm in unweighted
graph model
• Searching behavior is
customizable in terms of
interaction properties:
biological type, evidence
and etc.
• Future: Dijkstra’s algorithm
in weighted model
Case#3: Building sub-network over a group of user-
submitted nodes
Shortest path
search algorithm
was extended to
construct the sub-
network for a group
of genes
Connecting to Araport --- Step 1: REST web service
• We developed REST interface for each service in HRGRN
• Example: searching “unknown” relationship between genes (case #2)
– When user visit
http://plantgrn.noble.org/hrgrn/path?hasParams=T&node1=np02084&steps=3&node2
=np12356&pathalg=allSimplePaths&PPI_validated=T&GENEEXPREGU_validated=T
&COEXP_predicted=T&format=json
– Web site will generate JSON code as below instead of full HTML5 web page:

3. 3
• A python proxy script providing common interface for
upstream web services (Araport team developed)
• GitHub account
• https://www.araport.org/api-explorer
Connecting to Araport --- Step 2: proxy portal Connecting to Araport --- Step 3: Science App
https://www.araport.org/apps/eriksf/hrgrn-app
• Araport can generate
an individual-gene
centered subnetwork
based on REST web
service from HRGRN
• No user customization
panel
Suggestion for Araport
• Supporting more languages for Proxy
adapter, e.g., Java, JavaScript and Perl.
Acknowledgement
The Noble Foundation
• Patrick Zhao
• Tingsong Liu
• Jun Li
• Zhaohong Zhuang
• Junil Chang
• Wenchao Zhang
JCVI
• Irina Belyaeva
• Jason Miller
• Chris Town
• Vivek Krishnakumar
Open Source Community
• CentOS Linux
• Oracle Java
• Resin, Java web server
• Groovy, Java-based script language
• Neo4J, open source Graph database
• Cytoscape.js, HTML5 web front-end
• JQuery, JavaScript framework