Driving Drug Discovery Efficiencies in Life Sciences with AI – with Anne Phelan of BenevolentAI

Riya Pahuja

Riya covers B2B applications of machine learning for Emerj - across North America and the EU. She has previously worked with the Times of India Group, and as a journalist covering data analytics and AI. She resides in Toronto.

Driving Drug Discovery Efficiencies in Life Sciences with AI-1-min

According to research from the American Society for Biochemistry and Molecular Biology, nine out of 10 drugs fail to make it to market in the clinical trials process. Researchers in the cited studies found that 40-50% of these failures are due to a lack of clinical efficacy, meaning that the drug is not able to produce its intended effect in people, and 30% were due to unmanageable toxicity or side effects.

With that backdrop, the potential of AI to streamline longstanding problems and inefficiencies in drug discovery processes is immense. Life sciences enterprises that leverage AI can analyze large datasets like genomic data, literature, and existing studies to identify potential drug targets and predict drug properties like binding affinity, efficacy, and toxicity. While still an evolving array of capabilities across maturing use cases throughout the life sciences space, AI shows promise in reducing time and costs, increasing success rates, and enabling novel drug discoveries.

Emerj Senior Editor Matthew DeMello recently spoke with Anne Phelan, Chief Scientific Officer at BenevolentAI, for Emerj’s new ‘AI in Healthcare and Life Sciences’ podcast channel to discuss how AI-assisted drug discovery can help make better predictions on target identifications. 

The following analysis of their conversation examines three key insights for business leaders:

  • Utilizing AI for target identification: Analyzing vast datasets and complex relationships to help in target identification by making precise predictions about targets with therapeutic potential.
  • Patient stratification for tailored treatment: Identifying patient characteristics and traits to enable a more tailored approach for treatments that enhances clinical efficacy and improves treatment outcomes. 
  • Expanding therapeutic opportunities with AI: Leveraging data to draw parallels between different chronic inflammatory conditions to broaden the therapeutic opportunity space and enable the discovery of novel treatments, thereby addressing unmet medical needs and improving patient outcomes.

Listen to the full episode below:

Guest: Anne Phelan, Chief Scientific Officer and Cambridge Site Head, BenevolentAI

Expertise: Clinical Project Leadership, Drug discovery, Molecular Pharmacology

Brief Recognition: Anne has over 20 years of experience in pharmaceutical drug discovery. She has previously worked with Pfizer Neusentis, where she last served as the COO and Head of Pharmacology. She obtained her PhD in Genetics from the University of Liverpool.

Utilizing AI for Target Identification

Anne begins her podcast appearance by discussing the challenges and costs of drug discovery and development. She highlights that the process is lengthy, complex, and risky, leading to a high failure rate and significant expenses. She emphasizes the importance of selecting the right target for drug development from the beginning. If the initial target is incorrect, it significantly decreases the chances of success, regardless of how well the drug is designed and refined. 

She mentions that currently, there is a 90% failure rate in drug discovery, meaning that only a small percentage of drugs entering clinical trials successfully make it to launch. The high failure rate contributes to the exorbitant cost of drug development, with each successful drug costing around the often-cited figure of $2 billion as the expenses of failed attempts are carried forward.

Anne then discusses the immense complexity involved in drug discovery due to the vast number of genes in the human genome (around 25,000) that could potentially be targeted to treat various diseases. The scenario creates a daunting task for scientists, who must navigate the extensive pool of potential targets to find those with therapeutic relevance for specific diseases. She emphasizes that that these challenges are precisely where AI can be particularly beneficial.

Anne emphasizes that AI can analyze vast amounts of data and complex relationships among various factors, enabling it to identify critical patterns and insights that would be difficult for humans to discern. In drug discovery, AI can aid in target identification by efficiently sifting through large datasets and making precise predictions about which targets are most likely to have therapeutic potential.

Patient Stratification for Tailored Treatment

Anne also discusses the process of drug development. Once scientists have pinpointed a target, such as a specific gene or protein associated with a disease, they need to consider various factors before designing a drug to modulate that target.

First, they must decide whether to activate or inhibit the target to understand its expression in the human body. They also need to determine if the drug needs to penetrate the brain or be kept out of the brain for optimal effectiveness.

“So this is where there’s a whole additional suite of AI models and algorithms that can really help us to design the right molecule that is going to very selectively, very specifically target your gene of interest. So it is very much sequential, and the models, even though the AI has the capacity to reason over huge amounts of data, can also be very exquisitely specific in the kind of questions you can pose.”

–Anne Phelan, Chief Scientific Officer at BenevolentAI

Anne further explains two key areas where AI plays a significant role in drug discovery:

  • Design and optimization: In the initial phase, scientists utilize a range of tools and models to predict the properties of the drug being developed. These predictions include stability within the human body, pharmacological properties, and other relevant factors. It’s a multifaceted optimization process where scientists aim to tailor the drug to achieve desired characteristics. 
  • Patient Stratification: While patients may have the same diagnosis, their characteristics and traits vary significantly. AI helps identify and strategize patients based on these unique traits, allowing for a more tailored approach to treatment. By matching the right drug to the right target and patient subgroup, researchers can enhance the likelihood of clinical efficacy, leading to better treatment outcomes. Doing so creates a virtuous cycle where the precision of treatment increases, potentially improving patient outcomes.

Expanding Therapeutic Opportunities with AI

Next, Anne provides an example from her organization’s internal portfolio to illustrate how they used AI in drug discovery, specifically for treating Ulcerative Colitis (UC).

  • Data Querying: BenevelontAI utilized its algorithms to analyze many data sources, including genetics data, literature, and information on existing drugs. All these data points were amalgamated into a comprehensive database to inform their research.
  • Patient Stratification: Within the population of patients with UC, there are different subgroups based on the severity of the disease. Anne’s team focused on patients with moderate to severe UC and sought to identify characteristics specific to the subgroup.
  • Target Identification: Based on the analysis of patient data, their AI models proposed a protein called PDE10 (phosphodiesterase 10) as a potential therapeutic target for patients with moderate to severe UC. The target was not previously linked to UC in the literature, highlighting the unique insights provided by AI-driven analysis.
  • Experimental Testing: The proposed target was further investigated through experimental testing, leading to the development of a drug discovery program. The company subsequently selected a PDE10 inhibitor and progressed it into clinical trials to treat patients with UC.

Anne shares with Emerj’s audience that traditionally, patient stratification involves considering clinical diagnostic components such as the severity of symptoms or the extent of inflammation in a specific organ, like the colon. However, AI enables researchers to delve deeper into patient stratification by incorporating molecular patient signatures.

AI algorithms can analyze molecular-level data from patients, providing a more detailed understanding of why certain patients may experience severe symptoms. A more granular approach to patient stratification enhances researchers’ ability to tailor treatments to specific patient subgroups, potentially leading to more effective and personalized therapies.

Anne then mentions the limitations of current therapies for certain conditions, such as osteoarthritis, despite the availability of treatments that provide some benefit. However, she notes that these therapies often have drawbacks, such as wearing off over time or not adequately treating all patients:

“So obviously, for patients who suffer from osteoarthritis, it’s a chronic inflammatory condition. Can we draw parallels with other chronic inflammatory conditions? Can we broaden the therapeutic opportunity space so that we can find novel treatments for these diseases? I think, with the advent of AI, we’ll be looking at a greater diversity of targets and a greater opportunity space to try and get something that is precisely well suited to – not an individual patient as in one individual, but these clusters of patients that have the same characteristics. Then we can start to target them much more effectively because the AI gives us profound insights into the disease that they’re suffering from.”

– Anne Phelan, Chief Scientific Officer at BenevolentAI

Anne lastly highlights that with AI, researchers can analyze data with unprecedented precision and gain insights into the intricate mechanisms underlying diseases. These capabilities include understanding not just the symptoms but also the disregulation of cellular processes, down to the level of proteins and mutations. Such comprehensive understanding allows for a more nuanced approach to identifying therapeutic targets:

“Even if AI only manages to move the needle a little bit on each of these steps, we find better targets, we find better drugs, and we’re better able to choose the patients. When you’re looking at an industry that has a 90% failure rate, and this $2 billion cost, even if you can improve the probability of success incrementally on each step, then fundamentally, you’re improving the probability of creating a drug that has therapeutic potential for patients who have nothing.”

– Anne Phelan, Chief Scientific Officer at BenevolentAI

Subscribe