Drug repurposing, real world data and AI/ML: perspectives and opportunities

De-novo drug discovery and development pipelines are time-consuming and costly. It can take up to 17 years and cost more than $2 billion dollars from the discovery of a new drug until it reaches the market. Drug candidates have a high rate of failure, as only 10% of de-novo drugs put through clinical trials finally obtain market approval, with the highest rate of attrition occurring at phase I and II of clinical trials (assessing safety, tolerability, dosage, efficacy, and side effects).

Drug repurposing approaches aim to select drugs that are either already on the market, or that have passed safety assessments in humans, and that could be leveraged to treat another condition than the one they were initially developed for. The most important benefit is that it can deliver efficacious therapeutics to patients awaiting treatment much faster.

Another non-negligible benefit is that it can increase the return on investment for pharmaceutical companies trying to put a new medicine on the market, as compared to traditional drug discovery approaches. Using a drug repurposing approach can lower the costs down to less than half a billion dollars, which can greatly impact a company’s research and development (R&D) budget.

In-silico methods, machine learning and big data

While in the past successful drug repurposing examples were mainly serendipitous, nowadays drug repurposing is more methodological and mostly driven by in-silico techniques. Computational methods range from molecular docking and binding-site detection to pathway mapping, genetic associations and machine learning (ML).

Ezequiel Anokian, Bioinformatics Consultant in the Discovery and Translational Science (DTS) Consulting Services at Clarivate, described the families of methods in a recent webinar, with a particular focus on the rapidly evolving ML-based techniques.

“Broadly speaking, we can categorize ML-based drug repurposing methods into four classes: classical Machine Learning, Network-based, Matrix Factorization and Deep Learning”, says Anokian. “Of these, Deep Learning models have gained popularity during recent years”.

  • Classical Machine Learning methods train traditional models like Random Forest or Support Vector Machines based on the observed samples (e.g., drug-disease pairs) and features (drug, target, and disease properties).
  • Network-based approaches operate on data structured as a network (graph), where nodes are biological entities (e.g., drugs, diseases or proteins) and edges are direct or indirect connections between them. Random Walk is a famous algorithm that falls within this category.
  • Matrix Factorization models work by decomposing a matrix representing a network (e.g., adjacency matrix) into smaller matrices in such a way that the multiplication of these gives an approximation of the original matrix. This process allows to impute/predict missing links. Matrix Factorization algorithms are widely used in social networks and recommender systems.
  • Lastly, Deep Learning models are composed of many hidden layers that learn abstractions of the input data. An Encoder-Decoder model is commonly used in drug repurposing: the Encoder part encodes the input amino acid sequence of the target or the SMILES of the compound into an embedding space, and the Decoder part predicts binding score between the input target and drug.

“Computational drug repurposing tools heavily rely on the quality and consistency of the data,” explains Anokian. “At Clarivate, we leverage three proprietary databases in our indication expansion and target identification pipelines: Cortellis Drug Discovery IntelligenceTM to extract disease biomarkers information, MetaBaseTM for protein-protein interactions, and OFF-XTM for drug safety and adverse events information. These manually curated and regularly updated databases represent one of the main pillars of our powerful workflows.”

Additionally, Real-World Data (RWD) can complement traditional randomized controlled trials, with the added advantages of being quicker, cost-effective, having less strict inclusion criteria, larger sample sizes, and longer follow-up periods. RWD can greatly support drug repurposing in the field of rare diseases.

The regulatory landscape

When trying to repurpose an existing drug for a new indication, some of the administrative workload can be lightened. For example, an investigational new drug can usually be skipped, as drugs proven to be safe in human (i.e., that successfully passed phase I clinical trials) are typically prioritized in the process.

“However, drug repurposing faces multiple challenges in the regulatory area,” points out Martí Bernardo-Faura, Senior Bioinformatics Consultant at DTS Consulting Services in Clarivate. “For example, in the U.S., the FDA offers a period of 3 years of data exclusivity for a new application of a previously approved drug. Most of the time, this 3-year period is insufficient for a company to recover the investment made to repurpose a drug.” In addition, off‑label use of a repurposed generic drug may further devalue the product.

As regulatory agencies become increasingly open to the concept of drug repurposing, the expectation is that more streamlined and less constraining regulatory pathways may encourage R&D teams to invest in this area. This is particularly relevant in the challenge of developing new treatments for rare diseases, as less administrative burden and increased return on investment may encourage rare and orphan drug development initiatives.

New business models

During the webinar Anokian and Bernardo-Faura noted that the establishment of collaborative new business models could be key to increased drug repurposing. Clarivate leads two subscription-based pre-competitive consortia for Algorithm Benchmarking (ABC) and Computational Biology for Drug Discovery (CBDD), where Clarivate benchmarks and implements advanced state-of-the-art pathway and network analysis algorithms that have a direct application in drug discovery and development.

Collaborations between industry leaders and innovators are key to boosting and empowering drug discovery and development. One example is Clarivate’s recent work with the U.S. Biomedical Advanced Research and Development Authority (BARDA) for repurposing drugs to treat chlorine gas or sulfur mustard injuries.

To learn more about the methods, challenges and perspectives of drug repurposing, please watch our recent webinar: https://clarivate.com/lp/harnessing-drug-repurposing/