Graph partitioning algorithms for detecting functional module from yeast protein interaction network

• Main Author: Abdullah, Afnizanfaizal
• Format: Thesis
• Language: English
• Published: 2010
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/11570/1/AfnizanFaizalAbdullahMFSKSM2010.pdf; http://eprints.utm.my/id/eprint/11570/
• Institution: Universiti Teknologi Malaysia

Advances in high-throughput technologies have provided many opportunities for researchers to study and better understand the dynamic mechanisms of systems biology. These systems are frequently formed by a functional organisation of networks that recapitulate specific biological processes. Protein interaction networks contain sets of sub-networks called functional modules with highly interactive proteins that perform similar functions. Recently, many graph partitioning algorithms have been proposed for detecting these modules, focusing only on detecting highly interactive proteins and neglecting proteins participating in sparse interactions. Moreover, many algorithms do not consider the overlap among different modules when identifying proteins that perform more than one function. In this research, new graph partitioning algorithms called Reliable Local Dense Neighbourhood (RELODEN) and Overlap-RELODEN are proposed to detect modules that contain highly interactive proteins, while also considering proteins with sparse interaction and overlap between different modules. The algorithms are based on the clique finding approach, which searches local cliques of informative proteins and groups the cliques into larger sub-networks. Experimental analyses using budding yeast (Saccharomyces cerevisiae) protein interaction network have shown that the proposed algorithms have the capability of detecting modules that are significant to biological functions, and thus giving a higher accuracy performance compared with existing algorithms. Moreover, these algorithms have found several interactive proteins that have not been reported previously, and are able to potentially predict the functions of a number of uncategorised proteins.