For all the consumer-facing headlines of generative AI’s healthcare failures—some less serious, others more troubling—a contrast is unfolding in the greater pharma and biotech universe.
In that world, the tone and tenor of genAI couldn’t be more different. Stakeholders are optimistic and unbridled in their imaginings of what the technology can and could do for drug development. With excitement for things like virtual cells, RNA sequencing, and so-called “AI scientists,” there’s much hope in AI’s preclinical future.
Some even expect that by 2050, we may have many machines helping scientists make discoveries that could today win the Nobel Prize. In fact, it’s already happening.
“We’ve seen and been a part of such a rapid rate of computation and sophistication of these technologies in what’s really a comparatively short amount of time,” said Rory Kelleher, global business development lead for healthcare and life sciences at NVIDIA. “The industry is absolutely right to feel optimistic about what more we’ll see from biological and chemical models in the future.”
Despite that enthusiasm, Kelleher noted that some key factors could limit AI’s momentum and impact in the future. Will the industry address those challenges with effective solutions so that future machines can live up to what life sciences needs them to be?
A whole world of “what ifs?”
Unlike other sectors, pharma and medical product leaders expect the most pronounced impact of AI in their industry will be on product R&D. Many imagine the value potential (defined by revenue percentage) to reach as high as 25%, for that matter.
As far as what the future could look like, the industry has been either contemplating, passively experimenting with, or even actively working toward many potential “what ifs.”
Several use cases support the ability of automation to enable new efficiencies—such as virtual cells that could simulate behaviors (potentially replacing animal models in the future) or robotic labs that could autonomously scale up experiments and scenarios. Decoding complex protein structures that have long eluded researchers is another exciting area, as is simulating clinical trials so that preclinical terms can better predict candidate performance in development.
Use cases for AI have other goals, too—such as enabling molecular simulation at a supercomputing scale for organizations of all sizes, as well as creating ”alien” hypotheses that humans would not feasibly think of thanks to resources such as knowledge graphs.
As these goals materialize, reducing the high cost of drug discovery and development through the efficiency of AI is a main hoped-for outcome that could help more quality candidates ultimately reach the clinic.
“Collectively, these use cases signal a future where AI fundamentally transforms drug discovery—accelerating research, reducing costs, and enabling breakthroughs that were previously unimaginable,” Kelleher said.
Barriers to overcome
For all the exciting future use cases of AI in drug discovery, many barriers present challenges that could limit those impacts. Data scarcity—including a lack of insights across the laboratory bench and human healthcare—is a significant one, suggests Kelleher.
“Data scarcity in biopharma is a significant hurdle for AI-driven drug discovery, but by embracing collaborative data sharing, leveraging synthetic data generation, and utilizing federated learning, we can overcome these limitations and unlock new potential in medical research,” he said.
Along with those scarcity implications, there are also concerns about bias and ethics surrounding the data that does exist. Experts have expressed concerns about AI models picking up human biases, for example, raising questions about how these concerns might impact downstream efforts to diversify clinical research when, or if, candidates get to human trials.
“There is certainly an important role for diversity and in how effective these models can be in drug discovery,” Kelleher said. “Any model is highly informed by the data it’s being trained on. So those biases definitely get carried over into the model and that could be a limiting factor in the future.”
Data scarcity and completeness notwithstanding, another potential problem point is regulator-related, and in particular the speed of AI adoption by the FDA. In a recent FDA-penned paper from JAMA, authors suggest that the rapid pace of AI implementation puts an onus on multiple stakeholders to help assess quality “beyond the remit” of the agency.
“The Alliance for Artificial Intelligence in Healthcare (AAIH) just discussed this with government regulators, actually,” said Stacie Calad-Thomson, business development lead, healthcare and life sciences at NVIDIA and vice chair at AAIH. “The FDA will hold drug filings to the same standards as now, but expect the speed and volume of files to accelerate to a point where the FDA will be a bottleneck if they’re slower to adopt AI solutions that will help them more efficiently process the larger volume of applications.”
Solutions for the future
As stakeholders work to unleash the potential of AI for future drug discovery, they should be prioritizing a foundational and non-siloed approach, Kelleher said.
“We need public, rigorous benchmarking, evaluation, and validation to be made public for scientific review by the community rather than being contained in commercial research siloes,” he added.
NVIDIA experts add that investigating and supporting open-source data factories, pre-competitive biobanks, and federated learning networks are other potential opportunities, along with the need to support the infrastructural demands of synthetic data generation.
Separately but similarly, there is a need to support the responsible use of data and cautiously embrace the idea of AI scientists and agents removing burdens from human teams so that they can focus less on clerical work. Experts say these solutions can help make preclinical through clinical milestones more seamless and practical.
“By implementing these solutions, we won’t just overcome current barriers,” Kelleher said. “We’ll revolutionize drug discovery itself, ushering in an era where AI accelerates breakthroughs and brings unprecedented therapies to patients worldwide.”
There’s a lot of excitement ahead about AI’s applications in drug discovery, but very real challenges too. Learn how NVIDIA is helping overcome concerns surrounding data scarcity, diversity, and other barriers at NVIDIA Healthcare