Prediction of miRNA-disease associations with a vector space model

Abstract

MicroRNAs play critical roles in many physiological processes. Their dysregulations are also closely related to the development and progression of various human diseases, including cancer. Therefore, identifying new microRNAs that are associated with diseases contributes to a better understanding of pathogenicity mechanisms. MicroRNAs also represent a tremendous opportunity in biotechnology for early diagnosis. To date, several in silico methods have been developed to address the issue of microRNA-disease association prediction. However, these methods have various limitations. In this study, we investigate the hypothesis that information attached to miRNAs and diseases can be revealed by distributional semantics. Our basic approach is to represent distributional information on miRNAs and diseases in a high-dimensional vector space and to define associations between miRNAs and diseases in terms of their vector similarity. Cross validations performed on a dataset of known miRNA-disease associations demonstrate the excellent performance of our method. Moreover, the case study focused on breast cancer confirms the ability of our method to discover new disease-miRNA associations and to identify putative false associations reported in databases.

Figure 1. Univariate distributions of AUC scores.

The small vertical ticks displayed on the horizontal axis correspond to the AUC values obtained for each of the 83 diseases. The AUC values were divided into bins of equal size. The vertical rectangles represent the number of cases for which the AUC value belong to each bin.

Figure 2. ROC curves obtained at each testing set for Breast Neoplasms.

Each color line illustrates the performance of MiRAI on one partition of the data. The curve is created by plotting the true positive rate (TPR) against the false positive rate (FPR) for all threshold settings. The best possible prediction method would yield a point in the upper left corner of the ROC space, which would represent 100% sensitivity (no false negatives) and 100% specificity (no false positives). The diagonal line represents a worthless method that gives random results.

Figure 3. Illustration of miRNA-target associations.

(a) miRNA-target bipartite network that connects items from a set of miRNAs to items from a set of targets. (b) One-mode miRNA projection that contains only nodes from the set of miRNAs. Two nodes are connected when they have at least one common connection to the same target. (c) One-mode target projection that contains only nodes from the set of targets. Two nodes are connected when they have at least one common connection to the same miRNA. (d) Weighted miRNA-target bipartite network. The graph is constructed in three steps: First, the resources that are associated with each miRNA node in graph (a) flow to the nodes that represent the targets. This operation indicates that miRNA a receives half of the resources of target α (a = α/2), b receives half of the resources of α and a third of the ressources of β (b = α/2 + β/3), and so on. Second, new resources that are associated with target nodes flow back to the nodes that represent the miRNAs. In the example, α receives all of the resources that are associated with a and half of the resources that are associated with b. Node α is now weighted with α/2 + (α/2 + β/3)/2 or 3α/4 + β/6. Third, the weights are applied to the original miRNA-target network. In the new weighted graph, some of the connections that are present in the original graph appear with updated weights (black lines with weights specified in blue), while new connections are highlighted (dotted lines with weights specified in red).