Shafique Virani, chief corporate development officer and interim chief medical officer, Recursion, delves into digital transformation within drug discovery and the power of AI-enabled platforms in the future of therapeutics.
Key insights:
- AI and new computational tools are allowing us to create and analyse broad, high-dimensional datasets that can dramatically accelerate the pace and scale at which we understand and control biology and chemistry.
- By using machine learning algorithms to uncover relationships between the disease model and a library of compounds, led Recursion to identify a chemical entity that had a strong and specific effect in the context of APC mutant genes.
- Recursion created a series of new chemical entities (NCEs) allowed more understanding of a molecule’s interactions. As a result, they developed a lead candidate, REC-3964, non-antibiotic approach to treat C. diff.
Over the past decade, there has been an explosion of research in the field of artificial intelligence (AI)-enabled drug discovery. Both the number of AI-native companies and the size of their pipelines have increased significantly. According to the 2022 State of AI Report, there are now 18 investigational drugs in clinical studies from AI-native companies. Two years ago, there were none.
This wave of new therapeutics is just the beginning of the digital transformation of drug discovery, powered by a new data-driven approach. Similar transformations in biotech have occurred in the past; in the 1990s, multiple new tools like DNA sequencing, genotyping and high throughput automation converged to fuel a genomic revolution in medicine. Today, AI and new computational tools are allowing us to create and analyse broad, high-dimensional datasets that can dramatically accelerate the pace and scale at which we understand and control biology and chemistry.
This is important because biology is extraordinarily complex. Despite scientific progress, we still see an abysmal 90% failure rate across all clinical trials, driving the total cost to bring each new approved medicine to approximately $2 billion. If we can uncover novel insights about the way biological systems and chemistry interact, we’ll create better medicines.
Here are two recent examples coming out of our labs at Recursion: our investigational program for familial adenomatous polyposis (FAP) and a potential therapeutic agent for the treatment of clostridium difficile (C.diff) infection, both for which we initiated new clinical studies in September of this year. Biologically, these diseases are very different – one is a rare tumour syndrome, while the other is a common bacterial disease. But together they represent the power of our AI-enabled platform to discover novel insights across diverse therapeutic areas.
Uncovering novel biology for FAP
FAP is a rare hereditary cancer syndrome affecting 50,000 people in the US and EU. Patients develop hundreds or even thousands of polyps along their gastrointestinal tract throughout their lives, which have a high risk of becoming invasive cancers of the colon, stomach and other tissues. Many patients eventually undergo a colectomy to avoid the disease progressing into colorectal cancer, but the predisposition to tumours remains a major cause of death following surgery. There are no approved drug therapies to treat FAP.
While the downstream biology of FAP can be varied and complex, the cause of the disease is well understood: mutations in a gene called APC cause FAP. This provides a unique opportunity for Recursion’s phenomics platform because we have an anchoring point from which to start our search. We can model and manipulate the disease in a cellular context and use machine learning algorithms to uncover relationships between the disease model and a library of compounds, unconstrained by the bias of any existing target hypothesis.
We initiated a screen using cellular models of loss-of-function of the APC gene. This led us to identify a known chemical entity that had a very strong and specific effect in the context of APC mutants, but not in the context of other kinds of oncogenes or tumour suppressor diseases. This particular molecule was originally studied by another pharmaceutical company to potentially treat colorectal cancer, but it had failed to meet its endpoints in clinical trials.
There was, however, a small subset of patients who had responded in the trial, but the company couldn’t explain why. The data we gathered from our platform and subsequent validation studies led us to believe that those were patients whose cancer was driven by APC mutations.
Today, this molecule is known as REC-4881, and it’s being studied in a clinical trial that is actively enrolling patients with FAP who have previously undergone a colectomy/proctocolectomy. It has been granted Fast Track and Orphan Drug designations by the U.S. Food and Drug Administration, as well as Orphan Drug designation by the European Commission. If successful, it could be the first drug therapy to treat FAP.
Uncovering novel chemistry for the potential treatment of C.diff infection
Clostridium difficile (C. diff) is a common bacterial disease that impacts more than 730,000 people in the US and EU every year. The current standard of care to treat C.diff infection relies on the use of antibiotics, which can negatively impact the gut microbiome and even lead to recurrence of inflection, which happens in 20-30% of all patients, or more severe forms of the disease (i.e. Toxic Megacolon). These patients who experience recurrent infections represent an opportunity for new treatment options that may provide benefit beyond standard antibiotic regimens.
We began by exploring relationships across cellular models of C. diff infection, which we created using different toxins produced by the C. diff bacteria, and our compound library. Following an initial ‘hit’ in our platform, we launched a comprehensive medicinal chemistry effort to optimise the compound properties to be suitable for the intended use. We created a series of new chemical entities (NCEs) that we fed back into our platform to drive structure-activity relationship (SAR) testing and optimisation. This process allowed us to gain a more complete understanding of the molecule’s interactions with the target and other biological systems, helping us reduce off-target activity and drug-drug interactions.
As a result, our lead candidate, REC-3964, is a completely novel non-antibiotic, small molecule approach designed to selectively inhibit the toxin effects produced by C. diff in the gastrointestinal tract. It’s the first NCE discovered by our platform to enter clinical trials. We believe this molecule has the potential, when used as part of a treatment regimen, to prevent recurrent disease and/or other forms of C. diff infection.
This is just the beginning
FAP and C. diff infection are just two of many examples in which big data and AI are being used to power a digital transformation in drug discovery. The beauty of AI-enabled platforms and the fit-for-purpose datasets that fuel them is the relatability of data across disease models, experiment types and therapeutic areas, revealing the interconnectedness of human biology. As our dataset grows with more experiments and inputs, the number of insights we can generate expands exponentially. This allows us to explore biology and chemistry at an unprecedented scale, and ultimately, to power the future of medicine.