Background: The prediction of essential genes from molecular networks is a way to test the understanding of essentiality in the context of what is known about the network. However, the current knowledge on molecular network structures is incomplete yet, and consequently the strategies aimed to predict essential genes are prone to uncertain predictions. We propose that simultaneously evaluating different network structures and different algorithms representing gene essentiality (centrality measures) may identify essential genes in networks in a reliable fashion.
Results: By simultaneously analyzing 16 different centrality measures on 18 different reconstructed metabolic networks for Saccharomyces cerevisiae, we show that no single centrality measure identifies essential genes from these networks in a statistically significant way; however, the combination of at least 2 centrality measures achieves a reliable prediction of most but not all of the essential genes. No improvement is achieved in the prediction of essential genes when 3 or 4 centrality measures were combined.
Conclusion: The method reported here describes a reliable procedure to predict essential genes from molecular networks. Our results show that essential genes may be predicted only by combining centrality measures, revealing the complex nature of the function of essential genes.