Narrow Results

Network diffusion processes play an important role in solving the information overload problem. It has been shown that the diffusion-based recommendation methods have the advantage to generate both accurate and diverse recommendation items for online users. Despite that, numerous existing works consider the rating information as link weight or threshold to retain the useful links, few studies use the rating information to evaluate the recommendation results. In this paper, we measure the average rating of the recommended products, finding that diffusion-based recommendation methods have the risk of recommending low-rated products to users. In addition, we use the rating information to improve the network-based recommendation algorithms. The idea is to aggregate the diffusion results on multiple user-item bipartite networks each of which contains only links of certain ratings. By tuning the parameters, we find that the new method can sacrifice slightly the recommendation accuracy for improving the average rating of the recommended products.

Shortest path length methods are routinely used to validate whether genes of interest are functionally related to each other based on biological network information. However, the methods are computationally intensive, impeding extensive utilization of network information. In addition, non-weighted shortest path length approach, which is more frequently used, often treat all network connections equally without taking into account of confidence levels of the associations. On the other hand, graph-based information diffusion method, which employs both the presence and confidence weights of network edges, can efficiently explore large networks and has previously detected meaningful biological patterns. Therefore, in this study, we hypothesized that the graph-based information diffusion method could prioritize genes with relevant functions more efficiently and accurately than the shortest path length approaches. We demonstrated that the graph-based information diffusion method substantially differentiated not only genes participating in same biological pathways (p << 0.0001) but also genes associated with specific human drug-induced clinical symptoms (p << 0.0001) from random. Furthermore, the diffusion method prioritized these functionally related genes faster and more accurately than the shortest path length approaches (pathways: p = 2.7e-28, clinical symptoms: p = 0.032). These data show the graph-based information diffusion method can be routinely used for robust prioritization of functionally related genes, facilitating efficient network validation and hypothesis generation, especially for human phenotype-specific genes.

The emergence of drug resistance to traditional chemotherapy and newer targeted therapies in cancer patients is a major clinical challenge. Reactivation of the same or compensatory signaling pathways is a common class of drug resistance mechanisms. Employing drug combinations that inhibit multiple modules of reactivated signaling pathways is a promising strategy to overcome and prevent the onset of drug resistance. However, with thousands of available FDA-approved and investigational compounds, it is infeasible to experimentally screen millions of possible drug combinations with limited resources. Therefore, computational approaches are needed to constrain the search space and prioritize synergistic drug combinations for preclinical studies. In this study, we propose a novel approach for predicting drug combinations through investigating potential effects of drug targets on disease signaling network. We first construct a disease signaling network by integrating gene expression data with disease-associated driver genes. Individual drugs that can partially perturb the disease signaling network are then selected based on a drug-disease network “impact matrix”, which is calculated using network diffusion distance from drug targets to signaling network elements. The selected drugs are subsequently clustered into communities (subgroups), which are proposed to share similar mechanisms of action. Finally, drug combinations are ranked according to maximal impact on signaling sub-networks from distinct mechanism-based communities. Our method is advantageous compared to other approaches in that it does not require large amounts drug dose response data, drug-induced “omics” profiles or clinical efficacy data, which are not often readily available. We validate our approach using a BRAF-mutant melanoma signaling network and combinatorial in vitro drug screening data, and report drug combinations with diverse mechanisms of action and opportunities for drug repositioning.