Drug Repositioning Strategies in the AI Era


Despite large R&D resources and expenses, drug discovery has become unbearable timely-wise and financially-wise for large companies, resulting in reduced number of new drugs, even though unmet medical needs are at their highest.

To address these problems, the medical sector innovates by implementing alternative approaches to optimize key steps of the drug discovery pipeline. Hence drug repositioning acts as a well-known strategy to find alternative indications for drugs that have already undergone toxicology studies. These drugs have mostly failed in later development stages and repositioning could give them a new target without the expenses needed to discover a new drug.

Nevertheless, repositioning approved drugs or candidates remains a challenging task, but the large quantity of biological data available provide fresh opportunities to create computational strategies in order to decipher and identify potent new way to use these drugs.

In this article we will take a look at the different strategies that could be applied to make drug repositioning.


The last few decades gave rise to a new plethora of computational techniques to develop efficient repositioning strategies (also called network-based drug repositioning). From molecular docking to pharmacological knowledge graph using deep learning algorithm; a growing number of solutions are developed in order to find new repositioning.

The primary purpose of these techniques would be to process automatically the numerous observations in drug discovery.

The four main categories of repurposing approaches

Figure 1: Target-based repositioning schematic (1).

Repurposing approaches can be divided into four main categories: 1. target-based, 2. side-effect-based, 3. expression-based, and 4. similarity-based.

  1. A Target-based repositioning is done by exploiting the role of its targets in diseases. Therefore, given a pathology for which a list of relevant targets is known, through the use of drug-target databases, all the possible drugs acting on such a list are evaluated as candidates for repositioning (2).
  2. A Side-effect-based approach focuses on the idea that it can provide clues to new therapeutic applications (3). For example, patients with benign prostatic hyperplasia treated with finasteride showed unexpected hair growth.
  3. An Expression-based strategy looks at expression profiles that can provide details on cellular state in response to a biological perturbation (drug treatment or disease) without any prior knowledge (4). Moreover, expression profiles can give an unbiased view of the entire coding genome, limiting side effects. The key concept behind these techniques is called signature reversion or signature matching. A repositioning is performed if a drug-disease pairs has anti-correlated expression profiles. If a gene is perturbed as a result of a disease, a drug that pushes such a gene in the opposite direction could be a therapeutic.
  4. Finally, a Similarity-based repositioning exploites the guilt-by-association (GBA) principle, if two pathologies share at least one common treatment, then some non-shared medication might be therapeutic for both diseases (5-6). Drug similarity can be assessed using chemical structure or known targets or common side effects, while disease similarity can be defined, for example, using ontologies.

Network-based drug repositioning

In biology, the concept of interaction network is heavily used. Based on this concept, Network-based drug repositioning have been developed and can be grouped based on their main source of biological data: 1. (gene) regulatory networks, 2. metabolic networks, and 3. drug interaction networks. Additionally, integrated approaches, using multiple data sources simultaneously, can be added as a fourth category.

We will describe further the types of algorithms that can be encountered as network-based repositioning in another resource.


A missing piece in current approaches to drug repositioning concerns electronic health records (EHRs). They offer a promising resource but a need for standardization is still essential for these data. For example by analyzing EHRs, an observational study could be performed by extracting unexpected effects, leading to novel therapeutic indications for existing drugs.

Moreover, the vast scale of EHRs data could enable large number of parallel drug repositioning tests, without any need of recruiting specific patients. But, to date however, very few EHR-based repositioning studies have been published.


  1. Sawada, R., Iwata, H., Mizutani, S., Yamanishi, Y. Target-Based Drug Repositioning Using Large-Scale Chemical-Protein Interactome Data. J Chem Inf Model Dec 28;55(12):2717-30 (2015). doi: 10.1021/acs.jcim.5b00330.
  2. Chavali, A.K., Blazier, A.S., Tlaxca, J.L., Jensen P.A., Pearson R.D., Papin J.A. Metabolic network analysis predicts efficacy of FDA-approved drugs targeting the causative agent of a neglected tropical disease. BMC Syst Biol 6:27 (2012). doi:10.1186/1752-0509-6-27
  3. Yang, L., Agarwal, P. Systematic drug repositioning based on clinical side-effects. PLoS One 6:e28025 (2011). doi:10.1371/journal.pone.0028025
  4. Sirota, M., Dudley, J.T., Kim, J., Chiang, A.P., Morgan, A.A., Sweet-Cordero, A., Sage, J., Butte, A.J. Discovery and preclinical validation of drug indications using compendia of public gene expression data. Sci Transl Med 3:96ra77 (2011). doi:10.1126/scitranslmed.3001318 
  5. Chiang, A.P., Butte, A.J. Systematic evaluation of drug-disease relationships to identify leads for novel drug uses. Clin Pharmacol Ther 86:507–510 (2009). doi:10.1038/clpt.2009.103
  6. Zhang, P., Wang, F., Hu, J. Towards drug repositioning: a unified computational framework for integrating multiple aspects of drug similarity and disease similarity. AMIA Annu Symp Proc 2014:1258–1267 (2014). PMCID:PMC4419869

*Please see the application resources on our Access Hub, or contact us.