Flipping the (Drug Discovery) Problem on its Head with Polypharmacology

For context: medicines have traditionally been designed to target a single protein with high specificity. However, the complexities of biology and pharmacology have made such single-target, “magic bullet” drugs rare. As I wrote in an article published in Forbes in Aug 2019, single-target drug design is not adequate for complex polygenic diseases, including but not limited to CNS-related diseases, autoimmune diseases, along with many cancers. 

And, as Professor Ross Cagan stated in a seminal 2012 paper while he was at the Icahn School of Medicine and Mount Sinai University, “the complexity of cancer has led to recent interest in polypharmacological approaches for developing kinase-inhibitor drugs…” It is now well accepted that approximately 300 protein targets are predicted to interact with a single drug. Considering that unexpected side effects - many of which are associated with off-target interactions - are the main reason for the failure of candidate drug trials, single-target drugs that do not consider off-target interactions pose a significant risk to drug discovery and patients (Zhou et al. 2015). It’s clear that understanding a drug’s polypharmacology is necessary to develop effective medicines that mitigate downstream costs associated with off-target failures.

Let’s look at one of the oldest and most infamous cases of polypharmacology, about which we wrote in the May 2017 blog post. In the 1950s, Chemi Grünenthal marketed a small molecule drug, thalidomide, to relieve symptoms of nausea and vomiting associated with morning sickness during pregnancy. Hailed as a “wonder drug” for its powerful sedative properties, thalidomide was sold in 46 different countries. Soon after thalidomide’s release, disturbing reports linked to the drug emerged worldwide. Studies confirmed that the wonder drug caused many unfortunate adverse effects, the most severe being the malformation or absence of limbs (phocomelia) in newborns whose mothers had taken thalidomide during pregnancy. It is estimated that thalidomide affected over 10,000 infants worldwide during its brief availability, though this likely underestimates the full impact of thalidomide, given that stillbirths and miscarriages went not counted. 

Despite its infamous past, thalidomide came back into use when Dr. Jacob Sheskin, a practicing physician in Jerusalem, prescribed the drug as a sedative to a leprosy patient. Surprisingly, thalidomide cleared the painful inflammatory lesions under the skin caused by leprosy. Today, thalidomide, is not only approved by the FDA for the treatment of symptoms of leprosy but has also gained significant attention as chemotherapy against multiple myeloma and other cancers. Thalidomide’s interaction with certain proteins in the body may cause adverse effects during pregnancy, but its interaction with other proteins simultaneously confers a therapeutic benefit in cancers. This realization that one drug may interact with several targets underlies the emergence of a new paradigm in drug discovery known as polypharmacology.

This is why at Cyclica, since 2014, we’ve been talking about and implementing the concept of polypharmacology in virtually everything that we do from the perspective of drug design. Unlike traditional computational methods like target-centric virtual screening and structure-based drug design, or QSAR modeling, our efforts have explored the possibility of exploiting polypharmacology to design better drugs. Instead of incrementally improving on a traditional single-target drug design, we’ve flipped the problem on its head and come at it with a proteome-wide, polypharmacology-focused strategy. Even some of our peers in the space have begun to champion the concept of polypharmacology.  

Harnessing the polypharmacology of molecules in designing better drugs is Cyclica’s raison d'être; it’s what our platform is trained to do, and with time why and how predictions keep getting better. 

Thus, we built MatchMaker, our proprietary deep-learning engine that uses structural and experimental assay data to predict drug-target interaction proteome-wide. MatchMaker powers our drug discovery platform for multi-objective drug design hit expansion and off-target profiling for phenotypic screening target deconvolution.

In March 2020, Cyclica launched the COVID-19 Stimulus package, offering in-kind use of Cyclica’s proteome-wide, deep learning engine to academics working on coronavirus. Our work on COVID-19 led to the publication of PolypharmDB as a resource to identify potential repurposed drugs that can be used to treat those infected with COVID-19. Alongside other prominent Canadian research institutions, Cyclica was able to determine that Capmatinib, an FDA-approved drug for lung cancer, interacts with protein targets that are associated with COVID-19 and other coronaviruses. 

Our collaboration with AstraZeneca - a case study which was published in January 2021 -  accentuated the strength of MatchMaker in identifying on and off-target interactions compared to experimental techniques like thermal proteome profiling, at a fraction of the time and cost. 

As for our most recent polypharmacological quest, last July Google’s Deep Mind made available the source code of AlphaFold 2, and a compilation of protein structures of 20 species generated in collaboration with EMBL.

We immediately began integrating the new structures and, within only a couple of weeks, had ~200,000 non-redundant pockets from 17,712 proteins, roughly doubling the coverage in our database. We believe that MatchMaker represents the most extensive AI-augmented proteome-wide screening capability in the pharma industry. Here, you can learn more about our work with AlphaFold2 structures on our blog series

The examples provided above are just a snapshot of how Cyclica has - and continues to - pioneer work in computational polypharmacology. If you’re interested to learn more about how Cyclica is tackling polypharmacology, please contact us!