IBM and Moderna explore quantum computing and generative AI with new partnership

The scientific miracle that is mRNA vaccines could be advanced even further in the coming years, thanks to a new partnership announced today between IBM and Moderna to explore the potential of generative AI and quantum computing.

In the early days of the COVID-19 pandemic, there was a race to create the first effective vaccine to help limit the risk of infection to the global population. It’s a race that mRNA-based vaccines, including ones from Moderna and rival Pfizer, were able to win through the use of innovative technology. As the risk of further COVID variants and other potential viruses still exists, drugmakers are looking to find ways to accelerate processes and discover new lifesaving approaches faster than ever before. That’s what has inspired Moderna to partner with IBM to explore the potential of new technological approaches that can help to solve problems faster. 

Those new approaches include the use of a generative AI foundation model developed by IBM that could potentially help Moderna create a new class of mRNA vaccine that doesn’t have the same limitations as current vaccines. On a parallel track, Moderna is also looking at how quantum computing can be used to help solve problems that existing classical computers cannot.

IBM’s broad push into generative AI continues

The Moderna partnership fits into IBM’s larger strategy for enabling organizations across different industry verticals to benefit from foundation models.

>>Follow VentureBeat’s ongoing generative AI coverage<<

“We are using broadscale foundation models for different use cases going all the way from healthcare and finance to geospatial with NASA,” Payel Das, principal research staff member and manager of trusted AI at IBM Research, told VentureBeat.

IBM announced its partnership with NASA to help build AI foundation models to advance climate science earlier this year. The partnership is one of many that IBM has with industry and scientific organizations to build AI models that solve real-world problems. IBM has also developed its own supercomputer platform with the primary goal of building foundation models. The Vela supercomputer integrates x86 silicon, Nvidia GPUs and ethernet-based networking. 

The MoLFormer foundation model

IBM will be working with Moderna on a family of foundation models known as MoLFormer. Das explained these models learn from broadscale molecular datasets. She noted the models can be adapted to different use cases, tasks and applications, and work in a different way than traditional deep learning models.

With many common predictive AI deep learning models, Das said that the challenge is that they are usually based on a limited number of samples, often generated by trial and error. “Classic machine learning and deep learning models struggle when they are trying to learn from a small amount of labeled data,” she said. 

In contrast, foundation models like MoLFormer work differently by being grounded in the broadscale knowledge of chemicals. She noted that MoLFormer is a general-purpose model that has been trained on more than a billion chemicals. As such, MoLFormer doesn’t need to have access to the data labels to learn new things from. Rather, the model already understands the underlying chemistry to be able to learn.

“With this general-purpose model, we can adapt it to different downstream tasks and we see amazing performance,” Das said.

How a foundation model reduces bias to improve AI outcomes

The foundation model approach could well help to reduce bias as well, according to Das.

She noted that in life sciences data there can be a risk from cognitive bias, as well as limitations stemming from the experimental setup. With the foundation model approach, the AI will be training on Moderna’s data, but it won’t be limited to that data alone. In the MoLFormer case, there are more than 1 billion chemicals already in the model.

“We are going to see the benefit of using this sort of foundation AI model architecture for going beyond the inherent bias that’s in a specific dataset,” Das said. “That’s one of the reasons why Moderna was actually interested in the foundation model paradigm.”

The quantum future

On a parallel track, Moderna is working with IBM to explore how quantum computing could be applied to their specific use cases in the life sciences.

 “Quantum computing holds the promise of more accurately modeling attributes and behaviors of molecules, which fundamentally operate according to the principles of quantum mechanics,” Jeannette (Jamie) Garcia, senior research manager, quantum computational science at IBM, told VentureBeat. “Quantum computing, which uses quantum algorithms that can leverage entanglement between qubits, can better capture the behavior of molecular systems.”

Garcia noted that Moderna is now part of the IBM Quantum Accelerator program, which provides the opportunity for Moderna scientists to explore how to apply quantum computing toward the development of, for example, future mRNA medicines.

In terms of where the intersection is between Quantum computing and AI, that’s an area that is also still being explored. Garcia said Quantum computing will always require an approach that leverages elements of both classical and quantum computing. In areas where quantum computers are integrated with classical and cloud computing, she noted there can potentially be an intersection in which AI resides.

“We expect these pathways to intersect as they both continue to advance,” Garcia said. “From our research, we know that AI also has the potential to benefit from exploring computational spaces to uncover insights in datasets unattainable with classical computing.”