PMC

Identification of the long QT disease gene neighborhood. Gene products (11,090) in the global human interactome were ranked on the basis of their relative MFPT distance score with the 13 LQTS disease susceptibility genes as seed nodes. Selecting various score cutoffs allows neighborhoods of various sizes to be defined as shown in the nested circles. Leave-one-out (leave-seed-out) cross-validation results, shown in red on the left, demonstrates the rank of seed nodes if they were not included as seed nodes. Several other gene products associated with other arrhythmia syndromes or the pathogenesis of TdP, shown in blue on the right, fall within the LQTS neighborhood. The protein-protein interaction network formed with the interactions between the 13 seed nodes and top 42 ranked nodes is shown at the top, demonstrating the assignment of relative importance to nodes based on network topology.

Multiscale relationships between mutated genes and QT interval. LQTS can be described across several organizational scales. (A) Organ level: LQTS symptoms result from disruptions of blood flow as a result of electrical disturbances of the heart that are observed clinically as a prolonged QT interval on an ECG. (B) Cell level: Changes in the QT interval arise from changes in the myocyte action potential leading to an increase in action potential duration. Normal and prolonged action potentials are shown as solid and dashed lines, respectively. The currents responsible for each phase are labeled. (C) Molecular level: The genes for ion channels that give rise to these currents and the proteins known to modulate these currents in LQTS are listed in the table. Known mutations in these genes allow clinical genetic testing to diagnose risk of LQTS. Dashed lines in (A) highlight the QT interval and in (B) highlight the action potential duration that gives rise to the QT interval.

Distribution of fractional overlap of other subnetworks with the LQTS neighborhood. The LQTS disease gene neighborhood was compared to disease gene neighborhoods generated from various control seed lists. These seed lists consisted of 13 randomly selected nodes, 13 randomly selected nodes for which their degree of connectivity matched the LQTS seed list, or 13 randomly selected nodes with GO annotation for the terms membrane and cytoplasm in a proportion matching those in the LQTS seed lists. Four hundred and twenty-two seed lists, with variable numbers of seeds, were created from the OMIM database. Each disease-related seed list was used to generate a neighborhood by selecting all nodes that had a positive score after MFPT ranking. The distribution of fractional overlap with the LQTS neighborhood, number of nodes shared by both neighborhoods divided by total number of unique nodes in either neighborhood, is shown for each set of random networks. Dashed lines indicate where the overlap threshold below which 95% of the random networks in each distribution falls. The 20 diseases showing the most overlap with LQTS are labeled. The “long QT syndrome” network with 87% overlap is based on the 11 seed genes that were in OMIM.

The use of the LQTS neighborhood to classify and explain FDA adverse event reports. (A) ROC analysis tests whether the LQTS neighborhood is more effective than the random neighborhoods for predicting if a drug, not previously reported as causing LQTS, has the risk of the side effect. Insets represent AUCs with mean and SD shown for the random network distributions. The numbers above the dotted line in the insets is the fraction of each distribution that had an AUC greater than or equal to the AUC found using the LQTS neighborhood. The point on the ROC curves that represent the cutoff for the LQTS neighborhood is labeled with S > 0. (B) The table lists selected drugs that were not previously annotated as prolonging the QT interval or causing TdP, but are associated with FDA AERS event reports of QT prolongation or TdP. (C) Tracing back to the LQTS seed nodes (red) from two drugs (blue), the Src inhibitor dasatinib (a cancer drug) and loperamide (an antidiarrheal) that interacts with calmodulin (CALM1), generates a subnetwork that suggests different ways in which these drugs used to treat different pathophysiologies could affect QT intervals.

LQTS neighborhood rankings classify QT-prolonging drug targets and drugs. ROC curves for MFPT-ranked nodes using the LQTS seed list, nearest-neighbor expansion from the LQTS seed list (NN), and the mean ROC curves for the various random seed list distributions are shown. The point on the ROC curves that represents the cutoff for the LQTS neighborhood is labeled with S > 0. Insets represent AUCs with mean and SD shown for the random network distributions. The numbers above the dotted line in the insets is the fraction of each distribution that had an AUC greater than or equal to the AUC of the LQTS neighborhood. In all analyses, the LQTS MFPT ranking significantly outperformed all other distributions (P values are indicated in insets). (A) Classification of drug targets. For every possible cutoff, a true positive rate (the fraction of targets of QT-prolonging drugs that are in the indicated neighborhood) is plotted against the false-positive rate (the fraction of all other drug targets that are in the indicated neighborhood). The AUC represents the probability that a given target of a QT-prolonging drug will rank higher than a given drug target that is not targeted by a QT-prolonging drug. (B) Classification of drugs. For every possible cutoff, a true positive rate (the fraction of QT-prolonging drugs that have a target in the neighborhood) is plotted against the false-positive rate (the fraction of the remaining drugs that have a target in the neighborhood). The AUC represents the probability that QT-prolonging drugs target higher-ranked targets than other drugs not associated with QT events. (C) Classification of drugs excluding HERG, the target encoded by KCNH2. The AUC represents the probability that QT-prolonging drugs target higher-ranked targets (other than HERG) than other drugs not associated with QT events. (D) Classification of drug classes. ROC analysis was performed with drug classes defined by grouping all drugs that share a common highest-ranked target, on the basis of the targets’ module distance scores, and including only these targets in the rankings. The AUC represents the probability that a given drug class containing QT-prolonging drugs target higher-ranked targets than drug classes not associated with QT-prolonging drugs.