Background: Computing centrality is a foundational concept in social networking that involves finding the most "central" or important nodes. In some biological networks defining importance is difficult, which then creates challenges in finding an appropriate centrality algorithm.
Results: We instead generalize the results of any k centrality algorithms through our iterative algorithm MATRIA, producing a single ranked and unified set of central nodes. Through tests on three biological networks, we demonstrate evident and balanced correlations with the results of these k algorithms. We also improve its speed through GPU parallelism.
Conclusions: Our results show iteration to be a powerful technique that can eliminate spatial bias among central nodes, increasing the level of agreement between algorithms with various importance definitions. GPU parallelism improves speed and makes iteration a tractable problem for larger networks.
Keywords: Centrality; Graphics Processing Unit (GPU); Iteration; Networks.