Prediction of conditional gene essentiality through graph theoretical analysis of genome-wide functional linkages.

The genome of an organism characterizes the complete set of genes that it is capable of encoding. However, not all of the genes are transcribed and translated under any defined condition. The robustness that an organism exhibits to environmental perturbations is partly conferred by the genes that are constitutively expressed under all the conditions, and partly by a subset of genes that are induced under the defined conditions. The conditional importance of genes in conferring robustness can be understood in the context of the functional attributes of these genes and their correlations to the defined environmental conditions. However, a priori prediction of such genes for a given condition is yet not possible. We have attempted such predictions by integrating the available gene expression data with genome-wide functional linkages through the well known centrality-lethality correlations in graph theory. We make use of three distinct concepts of centrality, namely, degree, closeness and betweenness, which yield mutually complementary information. We then demonstrate the efficacy of combined graph theoretical and machine learning approaches in ranking essential nodes from a large network of genome-wide functional linkages, which yields predictions with high accuracy. We therefore perceive such predictions as highly useful in applications such as defining and prioritizing drug targets.