Finding hubs in brain networks

Background and Aim : Brain networks could be defined as graphs which contain a set of nodes (neural components) and the edges (the connections between them). Among the many nodes form a network, some play a crucial role in mediating a vast number of network connections (called hubs). Hubs are central in network organization and are often identified by quantities known as centrality metrics. These nodes should have strong connections with each other to control the information flow. In this abstract, we investigate the nodes in form of a set and decide whether a set of nodes can be hub or not by using genetic algorithm.Methods : The parts of a genetics algorithm, are the population and function for life. The length of chromosomes is equal to the number of the nodes in the network. Being 0 shows that its corresponding node isn’t a possible hub and being 1 shows that it is a possible hub. In fact, in this method, each chromosome presents a possible hub. The aim is to find a chromosome that its proposed nodes have the most and strongest connections with each other. The cost function being used in this analysis sum all the edges between the considered proposed nodes and specify the result as the value of that chromosome. The chromosome which has the most value in the last generation, is the answer to the problem. Two sets of data have been tested by this algorithm Cat Cortex (is weighted and consists of 52 zones and has 4 main clusters) and Macaque Cortex (consists of 47 zones and is non-weighted and has 3 main clusters).Results : We executed our algorithm on Cat network for 100 times. The highest gained value was 73 so that the related hub set was repeated for 7 times. The hubs being found, cover all the clusters. For all the solutions to the problem, Epp zone was recognized as hub, because this is the only node of the Auditory cluster that has connection with the nodes from other zones and the nodes like DLS were not found in the problem’s solutions since they were not connected to other clusters and such nodes are not our algorithm’s favorite nodes. We executed this algorithm on Macaque network for 100 times and gained the responses’ frequency. According to the results obtained from this DB, we sometimes observed that no Somatosensory was found, because in comparison to other clusters, this cluster is too small and its nodes don’t have connections that much.Conclusion : The more the difference rate in clusters’ size is, the more possibility to bypass the clusters with small size, because the big clusters have stronger nodes that causes the hub being chosen from them. The nodes with few cluster interconnections are not a good candidate for the hub and also the nodes with few cluster interconnections are mostly ignored. A convergence towards the optimized response is being observed throughout the algorithm and the similarity among the nodes being found for 100 times of repeating the algorithm was a lot.

کلیدواژه ها:

Brain network ، Hub ، Genetic algorithm ، Cat cortex ، Macaque cortex