PREDICT: a method for inferring novel drug indications with application to personalized medicine

Abstract

The authors present a new method, PREDICT, for the large-scale prediction of drug indications, and demonstrate its use on both approved drugs and novel molecules. They also provide a proof-of-concept for its potential utility in predicting patient-specific medications.

Abstract

• We present a novel method for the large-scale prediction of drug indications that can handle both approved drugs and novel molecules.

• Our method utilizes multiple drug–drug and disease–disease similarity measures for the prediction task, obtaining high specificity and sensitivity rates (AUC=0.9).

• Our drug repositioning predictions cover 27% of the indications currently tested on clinical trials ( P<2 × 10 ⁻²²⁰).

• We show comparable performance using a gene expression signature-based disease–disease similarity, laying the computational foundation for predicting patient-specific indications.

Abstract

Predicting indications for new molecules or finding alternative indications for approved drugs is a laborious and costly process ( DiMasi et al, 2003), calling for computational solutions that would minimize production time and development costs ( Terstappen and Reggiani, 2001). Here, we present a novel method for predicting drug indications, PREDICT, capable of handling both approved drugs and novel molecules. Our method is based on the assumption that similar drugs are indicated for similar diseases. To score a possible drug–disease association, we compute its similarity to known associations by combining drug–drug and disease–disease similarity computations. This strategy achieves high specificity and sensitivity rates in a cross-validation setting, where part of the known associations are hidden and the method is assessed based on how well it can retrieve them based on the rest of the associations. Assessing its predictions of novel indications for existing drugs, we find that it covers a significant portion (27%, P<2 × 10 ⁻²²⁰) of drug indications currently tested on clinical trials. Examples of such predictions include: (i) Cabergoline, indicated for Hyperprolactinemia, which is predicted to treat Migrane, a prediction supported by two separate studies ( Verhelst et al, 1999; Cavestro et al, 2006) and (ii) Progesterone, which is predicted to treat renal cell cancer, non-papillary (npRCC), supported by the study of Izumi et al (2007). In addition, we provide indication predictions for novel molecules. For example, Cycloleucine is predicted for the treatment of Alzheimer's disease (AD); indeed, Cycloleucine was found to be a potent and selective antagonist of NMDA receptor-mediated responses ( Hershkowitz and Rogawski, 1989), a new promising class of chemicals for the treatment of AD ( Farlow, 2004). As another example, Hyperforin, St John's wort extract, is predicted to treat hyperthermia. Interestingly, St John's wort extract was found to have anxiolytic effects on stress-induced hyperthermia in mice ( Grundmann et al, 2006). We further introduce a disease–disease similarity measure based on disease-specific gene signatures and show that such a measure can be used by our method to accurately predict drug indications. Importantly, this suggests the potential utility of our approach also in a personalized medicine setting, whereby future gene expression signatures from individual patients would replace these disease-specific signatures.

Abstract

Inferring potential drug indications, for either novel or approved drugs, is a key step in drug development. Previous computational methods in this domain have focused on either drug repositioning or matching drug and disease gene expression profiles. Here, we present a novel method for the large-scale prediction of drug indications (PREDICT) that can handle both approved drugs and novel molecules. Our method is based on the observation that similar drugs are indicated for similar diseases, and utilizes multiple drug–drug and disease–disease similarity measures for the prediction task. On cross-validation, it obtains high specificity and sensitivity (AUC=0.9) in predicting drug indications, surpassing existing methods. We validate our predictions by their overlap with drug indications that are currently under clinical trials, and by their agreement with tissue-specific expression information on the drug targets. We further show that disease-specific genetic signatures can be used to accurately predict drug indications for new diseases (AUC=0.92). This lays the computational foundation for future personalized drug treatments, where gene expression signatures from individual patients would replace the disease-specific signatures.