PAIR v3.5 Update

Network analysis tools are now online.

PAIR now provides tools to analyze the predicted Arabidopsis interactome. These tools, located on the “Analysis” page, works with the “guilt-by-association” strategy to predict the function of a gene or identify the biological processes that are functionally connected to a given set of genes. The usefulness of these analysis has been discussed in one of our previous publications.

On the “Analysis” page, users can search a GO term for other biologically related GO terms. These GO term linkages represents “functional relevance” or “biological relevance”, while the GO term structure represents “sematic similarity” between GO terms. Users can also specify a group of GO terms to query for the GO linkages involving them or between them. Furthermore, users can use an observed gene set to query for GO biological processes that are tightly connected to them. In profile comparison experiments such as microarray analysis, results are usually differentially regulated gene sets. This analysis can help to interpret the biological significance of these differentially regulated gene sets, and connect the biological process under study to the well-established GO biological processes. Finally, GO annotations are predicted for proteins based on the GO annotations of their neighboring proteins in the interaction network. These predicted annotations can be queried on the “Protein Function Search” page.

PAIR v3.4 Update

PAIR server is now built with the J2EE architecture.

J2EE architecture is more flexible and can seamlessly integrate web services. We have shown in a recent paper that PAIR can be used as a resource to support several network topology-based systems biology analyses, such as gene set linkage analysis, gene function prediction, and identification of potential regulatory genes showing insignificant expression changes. We are currently implementing integrated analysis tools for these systems biology analyses. This server architecture change is the first step.

This is only a software system update without any change of the data content. The next major release, which updates the predicted interactions, is scheduled at the end of this year.

PAIR v3.3 Update

New experimentally determined interactions and improvements in the network visualization tool.

In this update, PAIR shows experimentally reported interactions with their experimental evidences. The set of experimentally reported interactions in PAIR 3.3 is collected from the major interaction data repositories including IntAct, BioGrid, BIND, and TAIR, as of July 23, 2010. As shown in the Interaction Statistics page, PAIR now contains 5,990 experimentally reported interactions and 145494 predicted interactions supported by the PAIR V3 prediction model. 1,445 (24.12%) reported interactions overlap with the predicted ones. Altogether, PAIR presents 149,900 interactions involving 10,380 proteins.

The integrated graphical interaction network browser has also been updated. The graph style was re-designed to improve clarity. More graph manipulation options have been added to facilitated pathway mining. Double click on a node will bring up a window showing all interactions involving this protein and allow selection of some interactions to be added into this graph. Right click on an edge will bring up an option to add this interaction to “My collection”. The layout of the network graph can be changed by applying a number of layout algorithms or simply dragging the nodes. The “Current Selection Detail” button will show a summary of the selected node or edge. Click the “Info/Tips” button for details.

In addition, the PAIR query algorithm has been optimized to improve the performance of the website and its background database.

PAIR v3.2 Update

An interactive network visualizer was implemented with Cytoscape Web.

The well-known interaction network analysis tool Cytoscape released its web version. In this update, we used this tool to replace our simple interaction network visualizer in PAIR V2.2. Cytoscape Web is more stable and provides better functionality. Now in the "Protein Information" page, "Interaction Search Results" page, and "My Collection" page, Cytoscape Web is used to display interaction networks. Its usage is briefed in HELP/FAQ pages. In addition, we implemented an "exportation" function for Cytoscape Web so that the interaction network displayed can be exported in various data formats, including the Microsoft Excel format, Cytoscape SIF format, GraphML format, PSI 2.5 XML format and our PAIR XML format. These improvements require the support of javascripts and Adobe Flash Player.

PAIR v3.1 Update

More informative web pages.

A number of web pages have been redesigned to present more information in a nicer way. For example, in the "Interaction Information" page, the relationship between the GO annotations for two proteins is now presented in a graphical way. The sub-cullular localizations shared by two proteins are now explicitly given. The gene expression profiles for interacting proteins are now illustrated by charts, so that users can develop an intuitive idea on how they correlate under different types of perturbations (e.g. light, stress, pathogen, development and abiotic). These improvements require the support of javascripts and Adobe Flash Player.

PAIR v3.0 Update

New indirect evidences, updated training dataset, and increased prediction accuracy.

In this major data release, several improvements have been made in our interaction prediction method. First, in addition to the indirect evidences used in PAIR v2.1, we added two new types of indirect evidences, phylogenetic profiles and homologous interactions. Second, we now use the five well-organized AtGenExpress expression datasets (light, stress, pathogen, development and abiotic) to calculate our co-expression features. Third, we updated our gold standard dataset to include recently reported Arabidopsis interactions. These changes have been documented in the HELP/FAQ page.

PAIR v2.4 Update

Added BLAST sequence search and modified the "My Collection" service (Test Update).

We had implemented the BLAST sequence search function on the Ruby on Rails framework. This means that this feature is still in its beta test, and is not yet available in the public website. In addition, we had changed the data storage mode for "My Collection". Now, users no longer need to create an account and login to use "My Collection". Interactions in “My Collection” are now stored in local browser cookies. Note that the cookies (the interactions) will normally be erased as soon as the browser is closed.

PAIR v2.3 Update

Switched to the Ruby on Rails framework (Test Update).

In order to improve the productivity of web service development, we have decided to change the technological platform of the PAIR website from JavaServer Faces to Ruby on Rails. In the next public release, PAIR website will be powered by Nginx, a free, open-source, high-performance HTTP server. The backend database management system will remain to be Oracle 10g. This 2.3 update is an internal update to test the Ruby on Rails technology. It can only be accessed through intranet and is expected to produce the same search results as version 2.2.

PAIR v2.2 Update

An interactive network visualizer was implemented with Flash (Test Update).

With the latest Flash technology, we developed a light-weighted interaction network visualizer. This visualizer offers an interactive way to search PAIR and construct a customized pathway. It allows users to zoom in/out specific areas of the interaction network, manipulate layout arrangements, and intuitively add or remove specific interactions in the interaction network graph. This test update can only be accessed through intranet.

PAIR v2.1 Update

New function “My Collection” was added to facilitate pathway mining.

PAIR now allows registered users to store their interested interactions in a personalized storage place called “My Collection”. Interactions in “My Collection” can be collectively exported into a single data file in the HUPO-PSI format, which is readable by major interaction network analysis programs. User registration is free. However, to ensure resource efficiency, dormant user accounts without activity in one month will be purged. Please export interactions as soon as pathway mining finishes.

PAIR v2.0 Update

Increased prediction reliability and public access.

Based on PAIR v1.5, we tested a number of feature extraction and feature selection approaches and then selected fourteen features representing four types of indirect evidences to construct an accurate SVM model for interaction prediction. The generalizability of this model was assessed by predicting the new additions to the BioGrid interaction database. Our model could recognize ~28.91% of them, which is close to the expected sensitivity estimated during model construction. In addition, we made PAIR accessible from the Internet.

PAIR v1.5 Update

Use support vector machine to predict interactions.

A support vector machine (SVM) model was constructed to predict Arabidopsis interactions based on seventeen features representing five types of indirect evidences. The SVM approach showed better receiver operating characteristics (ROC) curve when compared with the naïve Bayesian approach, and therefore shall produce more accurate predictions. The reliability of predicted interactions was not yet rigorously assessed. Next update will provide relevant data.

PAIR v1.0 Update

Use naïve Bayesian approach to rank interaction reliability.

Following the idea of evidence integration (Science, 302:449-53), we computed seventeen features representing indirect evidences of protein interaction. A naïve Bayesian model was then constructed to rank the interactions predicted by homology as in PAIR v0.5.

PAIR v0.5

Homology mapping of known protein interactions to Arabidopsis.

To facilitate the molecular mechanism studies in Arabidopsis, we compiled a comprehensive list of experimentally reported high-confidence interactions in Homo , Saccharomyces cerevisiae, Caenorhabditis elegans and Drosophila melanogaster. These interactions were then mapped to Arabidopsis interactions based on the ortholog groups assigned by the InParanoid database.