Source Identification in H1N1 Flu Infection

Background. H1N1 is a worldwide epidemic, and targeting early stage of human proteins involved will be effective in treatments due to high mutability of H1N1. Aim. We aim to identify source proteins in H1N1 infection. Data. Human protein-gene network containing protein-protein, post-transnational modification and protein-DNA interactions forms a weighed directed graph given transmission time of each edge. At different time points of H1N1 infection, expressions of part of gene were recorded and activation time of these gene can be obtained based on these records. Method. Activation starts from source protein nodes and diffuses toward neighbors. Availability of activation time from part of gene nodes makes the problem become diffusion network with partially observed nodes. Starting with the shortest path tree from the original graph, we have developed two algorithms for source identification. The MLE with edge optimization performs iterative update of likelihood for each potential source node and identifies source based on likelihood ranking. The fastest diffusion method identifies source nodes which overall propagate activations towards observed nodes faster. Results. Both methods are better than random guess, and the fastest diffusion method gains about 12 folds of source nodes hits compared to random guess. Conclusions. Source proteins in H1N1 infection can be traced in the methods here with much better chance than random guess. The methods derived in the study can be applied not only to source prediction in H1N1 infection but also in other diseases.