Identifying serendipitous drug usages in patient forum data a feasibility study

Drug repositioning reduces safety risk and development cost, compared to developing new drugs. Computational approaches have examined biological, chemical, literature, and electronic health record data for systematic drug repositioning. In this work, we built an entire computational pipeline to investigate the feasibility of mining a new data source – the fast-growing online patient forum data for identifying and verifying drug-repositioning hypotheses. We curated a gold-standard dataset based on filtered drug reviews from WebMD. Among 15,714 sentences, 447 mentioned novel desirable drug usages that were not listed as known drug indications by WebMD and thus were defined as serendipitous drug usages. We then constructed 347 features using text-mining methods and drug knowledge. Finally we built SVM, random forest and AdaBoost.M1 classifiers and evaluated their classification performance. Our best model achieved an AUC score of 0.937 on the independent test dataset, with precision equal to 0.811 and recall equal to 0.476. It successfully predicted serendipitous drug usages, including metformin and bupropion for obesity, tramadol for depression and ondansetron for irritable bowel syndrome with diarrhea. Machine learning methods make this new data source feasible for studying drug repositioning. Our future efforts include constructing more informative features, developing more effective methods to handle imbalance data, and verifying prediction results using other existing methods.