AI in CAR T Development: Lessons From the Front Lines

Within the life sciences industry, companies are constantly experimenting with how to incorporate AI tools into their processes. Companies developing cell therapies, such as CAR T therapies are particularly interested in AI integration designed to lower the overall cost of drug development and manufacturing, translating into lower costs for both patients and payers while increasing patient access. However, navigation of the nuances and challenges of this integration can be daunting. BioSpace spoke with Alex Sargent, director of process development at Charles River Laboratories, and his colleague Alan Smith, executive director, global scientific portfolio management, CDMO, to understand the promises and practical challenges of integrating AI tools.

The data topic has been circulating for years, to the point that the phrase “garbage in, garbage out” induces eye rolls. However, data-focused conversations are still a necessity. Sargent emphasized that one of the most surprising discoveries for teams venturing into AI-enhanced CAR T therapy development has been the massive data requirements needed to build effective models. Sargent, who has been working extensively with machine learning (ML) applications in CAR T therapy processes, recounted data challenges as one of the first issues he encountered.

Because the initial ML models his department employed have been around for decades, he found that incorporating them into process design for CAR T therapies was easier than anticipated. “The challenging part, which I did not realize going into it, but I quickly found out, is how much data you need to feed a model or to create a model,” Sargent continued. He recalled having a conversation about using machine learning to analyze data in which he was told that a lot of data is needed. “I said, ‘Don’t worry, I have tons of data,’” Sargent recounted. “My collaborator said no, we want 10,000 data points. We could get away with about 1,000. That right there blew my mind.”

As the field advances, Smith said he anticipates that new challenges around data management and utilization will arise. However, he added that he sees both opportunity and complexity ahead for AI in CAR T therapy development.

“As more and more information becomes available and more is known about relationships and impact from the biology, I think one of the challenges is how do you manage the growth and available information. How do you effectively frame what period of data acquisition in your big data set is usable or most useful?” Smith observed.

While the industry is struggling with data challenges, technology is moving forward into quantum computing. With major players such as GSK and Microsoft making significant investments in the technology, it is beginning to capture the attention of CAR T therapy developers.

However, the approach to emerging technologies varies among companies and individuals, as demonstrated by Sargent and Smith. In a lab environment, Sargent is constantly evaluating new technologies to identify their value to the industry and to Charles River Laboratories. “From my standpoint, it is about evaluating anything and everything. It doesn’t mean that we’re going to adopt everything we get our hands on, but I certainly think that evaluating every system, every algorithm, and every advancement is worth our resources and our time. It is part of what makes a scientist a scientist.”

Smith, meanwhile, echoes the sentiments of the risk-adverse within the industry with his cautious approach. “I’m more a wait and watch guy where Sarge is up to his eyeballs in all of this stuff. I think you’ll see folks like Sarge who want to use and learn more about anything and everything that’s available, any tool that might be applicable.”

An important perspective emerging is the understanding that AI isn’t entirely new. While there still seems to be confusion about the history of AI within the industry, more scientists are recognizing that many AI applications have deeper roots than commonly assumed.

“I do think I’ve heard this argument that AI machine learning has been around for a long time. We’ve been using it for a long time, right? If you have a machine that analyzes an image and says this is this, and this is not this. That’s a classifier algorithm, you know, and that’s been available maybe even since when Alan was in grad school,” Sargent noted. Smith confirmed, “It goes back to the 50s even.”

According to Sargent, large language models (LLMs) are driving current excitement. “That is what has really been the most disruptive and the most talked about, even in the biological sciences, which is very interesting to me—the disruptive element that we’re trying to figure out how to use it and how much we can rely upon it.”

AI in biological research received significant validation when the 2024 Nobel Prize in Chemistry was awarded University of Washington’s David Baker, also of Howard Hughes Medical Institute, and Demis Hassabis and John Jumper, Google DeepMind for computational protein design. Sargent sees this as both recognition of AI’s potential and acknowledgment of its evolutionary rather than revolutionary nature.

“The Nobel Prize in Chemistry last year was for computational protein design. It was AI. This platform they had developed that could really harness and do powerful protein predictions on a scale never before seen,” he explained. Sargent was quick to point out that it was not reinventing the wheel. “We’ve been doing structural-based protein design and prediction for 60-ish years now. This is just potentially a better tool to do that.”

As CAR T therapy developers continue to navigate the AI landscape, Smith anticipates a period of rapid evolution full of growing pains. “I think there’ll be some areas where there will be interesting advancements. I also think that there will be some mistakes being made or we end up having a retrench after an application doesn’t work as intended.”

As the field continues to mature, the experiences of early adopters will likely shape how AI transforms cellular therapy development for years to come.

This article was written in partnership with Charles River Laboratories.