Hopper Fellow Applies Innovation, Collaboration to Foundation Models

The 2025 Admiral Grace Hopper Fellow, Alex Morehead, arrived at Lawrence Berkeley National Laboratory (Berkeley Lab) this summer with a vision: to combine his twin passions for bioinformatics and machine learning in ways that would benefit the whole scientific community. Now he’s leaning into what Berkeley Lab does best – team science and innovation – to make that vision a reality.

The Hopper Fellowship was established in 2015 to support young computational scientists in making outstanding contributions in high performance computing (HPC). Fellows spend two to three years at Berkeley Lab applying their skills in advanced algorithms, software techniques, HPC systems, and networking for scientific discovery. For Morehead, the technical and human benefits of bringing his research to Berkeley through the fellowship were a no-brainer.

“What drew me to the Hopper Fellowship is the freedom to work on open-source research in a very well-equipped setting, not just intellectually, but also in terms of resources, especially computing resources to run the experiments I want to run,” he said. “There wasn't any other place I could think of that would let me do that—give me all the compute resources I need to run interesting, exciting experiments, but also the people I need to collaborate with to design those experiments.” He’s using HPC resources at the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy (DOE) user facility located at Berkeley Lab, to power his research.

Morehead’s goal is to build a type of foundation model that would go beyond predicting structures of existing molecules and be able to generate new ones, a tool that could be game-changing for researchers in both chemistry and biology. “I'm interested in working in both domains at the same time, building a single type of machine learning model that can generalize across tasks that chemists and biologists care about,” said Morehead. “Whether it's predicting properties of molecules or generating new molecules or proteins from scratch, ideally they would be able to do that with a single type of machine learning model rather than a bunch of separate specialist models.”

To gain broad perspectives on how to make such a model happen in a data-efficient manner, he’s been collaborating with Berkeley Lab machine-learning specialists, learning about ways to incorporate physical principles into neural networks and increase their efficiency in tackling complex datasets in biology and chemistry.

It’s this culture of cross-pollination that Morehead sees as one key to success. “The biggest focus is just finding like-minded people who are interested in collaborating, particularly on open-source, open-science research where the expectation is not necessarily to commercialize, but to open source data and source code at the end of a project and try to speed up science,” he said.

Computing and expertise at NERSC will strengthen those partnerships; in addition to the power of the Perlmutter supercomputer, Morehead has access to the HPC strategy, skills, and infrastructure he’ll need to produce and refine his models. He said, “NERSC is in a great position to support this type of open project – it has the computing resources to run the experiments and to facilitate those collaborations.”

For NERSC’s part, supporting large-scale machine-learning research to benefit science is a key part of the mission. “We’re thrilled to have Alex working with us as a Hopper Fellow. His work will push on the frontiers of foundation models for science in computational biology,” said Wahid Bhimji, NERSC’s Division Deputy for AI and Science. “This kind of research requires cutting-edge, large-scale computing resources and is a great fit for our supercomputers at NERSC.”

In choosing to apply for the Hopper Fellowship, Morehead said Berkeley Lab’s reputation preceded it. He’s excited to take his place among those moving science forward in Berkeley.

“I think it goes without saying that the Lab has a great reputation in the field of open science and open source research for just machine learning and computing science in general,” he said. “That was also a big thing that drew me to it—there's a rich history of people who've come out of these kinds of fellowships and out of Berkeley Lab who have done some very inspirational work. That was a big thing that drove me to this position: just seeing the track record of the Lab.”