Link prediction

Andrew was involved in a link prediction problem study with recent Odum graduate, Tad Dallas and colleague John Drake, which was recently published in PLoS Computational Biology. They wanted to (i) estimate missing but likely links in a host-parasite network, (ii) test if those missing links changed the network structure and (iii) for a real data set, establish what biological features best predict missing links. They were able to recover randomly removed missing links with high accuracy, and showed that putting these links in to the host-parasite network can dramatically change network structure. For a rodent-parasite dataset, they found that host litter size and diet breadth, along with parasite taxonomy were important features for accurately predicting missing links.