Mathematical network analysis has been proven to be a useful and powerful tool for biological networks including networks of protein interactions, gene similarity and metabolic interactions.
Here I use network analysis to model human cancer and predict which genes or reactions are essential for cancer to allow it to grow or recovery from stress. A general assumption for biological networks is that the centrality of a node is in some way reflective of its biological importance. So I evaluated a wide range of weighted and unweighted node centralities and measures derived from centralities to predict reaction essentiality in metabolic networks. The metabolic networks are Mass Flow Graphs (MFGs), based on the Recon2 reconstruction of human metabolism and constrains on reaction fluxes from the PRIME algorithm. The edge weights in the networks are computed from Flux Balance Analysis (FBA) results in a selection of human cancer cell lines from NCI-60. I could not detect a direct connection between node essentiality and any centrality, but there is a correlation between the overall change of the centrality distribution in the inhibited condition compared to wild type and the inhibited reaction essentiality. With this I have found a promising network measure that can be used to predict possible drug targets. With MFGs a wide range of cellular conditions can be modelled, but only when we know what the cellular objective for FBA is.
When cancer cells are put under stress through treatment, they adapt their metabolism to react to the stress. This dynamic process with fluctuating gene expression is difficult to capture in a metabolic network. A better way to analyse the recovery process is to extract which genes are active during which phase. I evaluated seven time points of gene expression data for multiple myeloma cells that were treated with a proteasome inhibitor (PI), which disrupts the protein recycling process. From the pairwise gene expression similarity I constructed network to cluster the genes into groups that are active at the same time. The networks were clustered with a random walk algorithm called Markov Stability and evaluated with gene enrichment analysis. From the resulting clusters, collaborators were able to extract tRNAs that activate a protein called GCN2 that is essential for recovery. Followup experiments showed that a combination of PI and GCN2 is lethal for multiple myeloma as well as a few other cancer cells.