AI is already helping many of us with things we do every day. It’s also fueling breakthroughs in research that hold promise of reshaping things on a global scale, like the discovery of new materials and improving medical care.
Two new research papers published this week in scientific journals, one in Nature and one in Nature Machine Intelligence, show how generative AI foundation models can exponentially speed up scientific discovery of new materials and help doctors access and analyze radiology results faster.
The research and its potential are the result of extensive collaborations among Microsoft, academia and the private sector. Working with its partners around the globe, Microsoft Research has developed generative AI foundation models — large-scale models that leverage advances in AI — focused on materials discovery and radiology. The models were built from the ground up on Microsoft Azure and are being shared publicly to speed up development and potential uses.
“Science may be the most important application of AI. At Microsoft, we believe that the ability of generative AI to learn the language of humans is equally matched by its ability to learn the language of nature — including molecules, crystals, genomes and proteins,” says Chris Bishop, director of Microsoft Research AI for Science. “It will allow us to harness AI for tackling humanity’s most pressing challenges, from sustainability to drug discovery.”
MatterGen: A key to discovering better materials — and solutions — faster
The development of new materials has been an unsung hero of human advancement. Think of how steel girders are the backbone of modern cities and silicon chips power smartphones. It is a painstaking and expensive process, akin to finding a needle in a haystack, and it can cost millions or billions of dollars.
That’s because developing new materials traditionally requires screening potentially millions of possibilities, a process that can take years with no guarantee of success. The latest research ushers in a new approach represented by MatterGen, a generative AI model that operates similarly to text-to-image and text-to-video AI models. Instead of screening a universe of possible materials, researchers propose specific properties and MatterGen generates new materials based on those properties.
Experiments have begun validating the concept. When a material generated by MatterGen was synthesized, its properties were within 20% of the targeted material’s properties.
Similar to generative AI’s impact on drug discovery, MatterGen will have a profound impact on how, and how quickly, a wide range of materials are designed in fields including electronics, energy storage and biomedical engineering.
Developing new efficient battery material, for example, could unlock more sustainable energy storage, while advances in superconductors could lead to groundbreaking improvements in medical imaging or quantum computing.
“From an industrial perspective, the potential here is enormous,” says Tian Xie, principal research manager at AI for Science Cambridge in the United Kingdom. “Human civilization has always depended on material innovations. If we can use generative AI to make materials design more efficient, it could accelerate progress in industries like energy, healthcare and beyond.”
RAD-DINO: Faster data for doctors, better care for patients
The second research breakthrough will help doctors get better and more comprehensive medical data faster, potentially speeding up diagnoses and improving patient care.
Mayo Clinic and Microsoft Research are collaborating to develop multimodal foundation models that integrate text and images for radiology applications. Initially, the teams are exploring the use of Microsoft Research’s AI technology with Mayo Clinic’s X-ray data. Part of that work involves research published this week referred to as RAD-DINO, named for its focus on radiology and a specific learning method. This new approach to improving imaging can help personalize patient care and improve diagnostic accuracy.
The goal is to give clinicians quicker access to information they need to treat patients, with initial efforts aimed at developing a model that automatically generates reports, evaluates how tubes and lines have been placed through chest X-rays and detects changes from prior images. This could improve how clinicians work and care for patients by providing more efficient and comprehensive analyses of X-rays.
“I am excited to share our collaboration with Mayo Clinic, one of the world’s leading hospitals, to tackle one of the most pressing challenges in healthcare: delivering faster and more precise medicine,” says Javier Alvarez Valle, senior director of Multimodal AI, Microsoft Health Futures UK. “A key hurdle lies in safely integrating generative AI into clinical workflows, and our work brings together the best experts from AI and medicine to make it happen.”
Top photo: A researcher works on a MatterGen generated material synthesized in lab. (Photo provided by Shenzhen Institute of Advanced Technology)