Combing social networks and pharmaceutical databases using heterogeneous network mining allowed a team from Drexel University to better identify therapies suitable for drug repositioning, according to work published in Artificial Intelligence in Medicine.
In their study, Christopher C. Yang, PhD, and Mengnan Zhao, a PhD candidate, sought to streamline the drug repositioning process, which has contributed to around 30 percent of FDA-approved drugs in recent years. Repositioned medications—existing therapies that have been rebranded for different indications—accounted for 20 percent of new drugs brought to market in 2013.
“The number of newly developed drugs that can enter preclinical tests and clinical trials has gradually declined, and the number of newly approved drugs has not kept up with the consistent increases in pharmaceutical research and development spending,” the authors wrote. “In light of these challenges, drug repositioning receives increasing attention from both academia and pharmaceutical companies, becoming an alternative and promising way for drug development.”
In contrast, de novo drug development still has a success rate of less than 10 percent, they said. Experts estimate it takes some 10 to 12 years to develop a drug from scratch, costing companies an average of $1.2 billion before the product ever makes it to market. Repositioned drugs, on the other hand, have already been validated by pharmaceutical and toxicological tests and could take as few as three years to reach the same point.
Current computational methods for drug repositioning exploit extensive knowledge of individual diseases and drugs’ chemical properties to match existing therapies to newer indications, Yang and Zhao said, but they took a different approach. Instead of employing traditional methods, they based their repositioning strategy on adverse drug reactions (ADRs).
“The rationale for an ADR-based drug repositioning approach is ADR and disease are both behavioral or physiological changes in response to drug treatment, and if drugs treating a disease share the same ADR, that ADR may serve as a phenotypic ‘biomarker’ for the disease,” the authors said.
Using ADRs as their intermediate, Yang and Zhao constructed a heterogeneous health network containing drugs, diseases and ADRs, and developed path-based heterogeneous network mining approaches for drug repositioning. They pulled data from three sources, including two pharmaceutical databases (PharmGKB and SIDER) and one social networking site (MedHelp).
Their assessment found combining data from PharmKGB and MedHelp identified 479 repositioning drugs—more than any alternative methods. In addition, 31 percent of those drugs were supported by evidence from PubMed.
The pair of researchers presented their findings to three medical experts, who concluded the repositioning drugs recommended by Yang and Zhao’s proposed algorithms were “very helpful” for them to narrow down potentially applicable therapies for certain diseases. They also said the experiment helped them understand the limitations of data sources we so often rely on.
“The results of drug repositioning are usually considered as suggestions, predictions or recommendations, not the results that can be approved for patients immediately,” Yang and Zhao said. “The main contribution of our work and many repositioning studies is suggesting novel drug uses for pharmaceutical companies and medical associations to conduct further in vitro and in vivo tests, effectiveness and risk evaluation, and clinical trials.
“Besides, our medical experts also said they would like to see such findings and explore whether there are off-label use opportunities from the repositioning drugs.”