Cell phenotypes are coordinated by complex networks consisting of thousands of genes and their products. Therefore, understanding this organization is crucial to elucidate cellular activities. Recently with the advent of high-throughput technology, many researchers have tried to construct the networks from microarray. The requirements for network inference have been surveyed as follows. Firstly, large-scale genes should be supported while false positives with insufficient amounts of samples should be alleviated. Secondly, Fast analysis has to be supported because inference algorithm needs a large amount of computations and takes much time for the inference. Lastly, not only algorithm itself but also the software should be provided, and the software should have extensibility for a variety of purposes of the network inference.
We have developed a new system for inferring genetic interactions satisfying the requirements above. To meet the first requirement, we have modified and utilized MONET algorithm which had already been developed. Secondly in order to support fast analysis, we have implemented parallel Bayesian network learning and employed a supercomputer as a computing server. For the third requirement, we have developed the software system which has extensibility in terms of use of servers, types of organisms, and functionalities. For the extensibility of use of servers, we have implemented three-tier system architecture, enabling us to append new servers or substitute existing servers, and for the extensibility of types of organisms we have constructed organism knowledge bases, making it possible to append and analyze other organisms additionally. Besides in order to supply the extensibility of functionalities by integrating existing analysis tools, we have developed the proposed system as a plugin of Cytoscape software which has been famous as a visualization tool for biological interaction networks, and we have witnessed that it is straightforward to ...