A network approach to improve link prediction in protein protein interaction network and its application in healthcare

Abstract

Proteins are chief actors which facilitate most biological processes in a cell. But they rarely act alone; rather they interact with each other to perform its functions. There are multiple methods to discover these interactions. But experimental discovery of these interactions are labor intensive and costly. Hence research community is now looking forward to computational methods which can make accurate prediction from protein-protein interaction (PPI) data, as it is more reasonable rather than blindly checking all interactions. newlineRecent advances in high throughput techniques have dramatically increased the availability of PPI data. But high level of noise, sparseness and skewed degree distribution of PPI networks is always a challenge for computational methods. Studies were conducted on various types of algorithms used to identify missing and spurious interactions from PPI network. Two new algorithms AGASEED and AGAWEED were designed to identify missing and spurious interactions which in turn lead to the design of AGAEED, an algorithm to remove entire noise from the PPI network. The main highlight of these algorithms is in its ability to predict interactions directly from the topological properties, without depending on any physical property of the network. This make the algorithm easily portable to link prediction problem in other complex networks like newlinesocial network, co-authorship network, etc. Detailed analysis and comparison with the existing popular schemes were carried out to establish the superiority of the proposed method. newlineMany complex networks in nature, like protein network, metabolic network, social network, citation network etc exhibit a mesoscopic level of organization, with groups of nodes forming tightly connected units, called communities, that are only weakly connected to each other.The thesis include a detailed study on existing community detection algorithms and by making use of perceptions derived from studies, a new algorithm CDSOM was designed. newline