AI-Enabled Clinical Trials: A Multiomic Roadmap to Success

Image Credit: © Kaikoro - stock.adobe.com

Conversations around the impact of artificial intelligence (AI) in the pharmaceutical industry seem to be everywhere, with AI often hailed as the solution to the perennial problem of the exorbitant cost to develop a new medicine. In these discussions, AI is often portrayed as a tool to improve the speed of compound identification. But while others debate whether the value of investments flowing into companies purporting to apply AI to drug discovery and development, overlooked is another area where AI can drive critical impact in drug development: designing more successful clinical trials.

Patients participating in clinical trials give us a lot—their time, their bodies, samples of blood, urine, or other biological samples for biobanking or further study, and most of all, their hope and selfless dedication to advancing science in order to help their fellow man. We have a responsibility to ensure that their sacrifices are not in vain. AI, combined with multiomic analysis of all of our patient samples, can help the industry make the most of data from early-stage trials, thereby improving the chances of success in future clinical studies by targeting the right patients and protocols. Using this approach, our company has run three successful trials in oncology and two in rare disease with zero failures. Here’s how we used our AI model to achieve this unusual success and how other companies can leverage AI models to improve their clinical trials too.

Understanding the mechanism of action and selecting the right patients

Truly understanding a compound’s mechanism of action can help a trial sponsor select the right trial design and target the ideal patient population, yet AI is often overlooked as a tool for gaining these insights. Using multiomic data from tissue, blood, and urine samples collected from patients in a Phase I trial, AI models can help trial sponsors map patient biology and better understand the mechanism of action and safety—not just for the population as a whole, but also how the drug interacts with patients with specific germline or somatic genomic profiles.

As an example, we tested a drug compound with a broad mechanism that targeted the mitochondria in a large Phase I trial with 104 patients with different cancers. Leveraging our multiomic data we could better understand how our compound induced a shift in metabolism and see that its effect was limited to those with more aggressive tumors. Knowing this helped us prioritize future studies to more narrow clinical indications, select patients with these types of cancer, and better understand the pharmacodynamic biomarkers predictive of success before beginning Phase II studies.

Leveraging safety and toxicity insight to design safer trials

AI models can also help companies understand the relationship between patient biology and adverse events and modify clinical protocols to limit toxicities and optimize trials for patient safety. Such modeling can help the sponsor understand why adverse events occur, predict what types of patients might be vulnerable to similar events in the future, or identify mitigation measures to reduce the potential for adverse events, either by changing the trial protocols or inclusion/exclusion criteria.

In one of our trials, our AI modeling identified toxicity markers for patients with previously unknown bleeding and clotting abnormalities leading us to add dosing of vitamin K to the clinical trial protocol in a subsequent trial.

Optimizing combination therapies and treatment approaches

Therapeutic combinations are becoming increasingly common, especially in cancer, but it can be difficult to know whether combination therapies should be delivered in parallel or in sequence, and in what order. AI can model each scenario a company would want to test, to identify the optimal protocol prior to clinical testing or to consider the best dosing regimen.

In cancer studies with multiple therapies or therapeutic modalities, AI modeling has helped us understand the effects of one protocol on other protocols, helping us narrow the focus of our study on how our product can enhance the effects of other therapies and radiation by dosing it ahead of other interventions.

Modeling continuous learning

When clinical trials aren’t successful, it’s just as important to understand why the protocol didn’t work in the patients who received it as it is to understand why a trial is successful. AI modeling is a continuous process of testing and learning and iteration, so that learnings about what’s not successful, can undergird future successes. AI makes this possible by helping researchers better understand a compound’s mechanism of action, and its effects on safety, diverse patient biology, or companion treatment approaches.

Niven Narain, PhD, President & CEO, BPGBio