The Eric and Wendy Schmidt Center (EWSC) at the Broad Institute of MIT and Harvard has chosen SEAS graduate student and Hekstra Lab member Minhuan Li as one of their inaugural EWSC fellows. EWSC and its fellowship support researchers who use machine learning to address questions in the biological sciences.
“I am grateful and excited to have been selected as a recipient of the EWSC Fellowship,” Li says. “The recently established center is working to build an innovative research community at the intersection of biology and machine learning, and I am delighted to be a part of this remarkable endeavor. I look forward to engaging in conversations with fellows from various backgrounds.”
In March 2020, Li joined the Hekstra Lab, which specializes in understanding protein structures and how they move. “It was an interesting time,” recalls MCB Professor Doeke Hekstra. “Most of the work in our group is focused on how to experimentally study the internal motions of proteins—the changes in shape, or conformation, that enable proteins to play their many roles in catalysis, transport, and assembly of dynamic structures. We mostly do so by collecting stroboscopic X-ray data on protein crystals as we perturb them. There’s still a lot to do on how we improve these data. When Minhuan joined right at the start of the pandemic, though, we couldn’t do experiments and decided to skip ahead to the problem of how we can fit trajectories, or movies, of how proteins change, to such data.”
Li’s thesis work focuses on the latter part of the protein analysis pipeline. He will combine x-ray crystallography data and deep learning algorithms to validate atomic models and protein trajectories against x-ray data. “X-ray data provides rich yet incomplete information about macromolecule structures, due to the so-called ‘phase problem,’” Li explains. “The current X-ray data processing pipeline still requires a lot of human intuition, especially at the final stage of structure refinement.”
Li will apply generative models to these problems of zeroing in on dynamic protein structures. “I have developed a differentiable pipeline between macromolecular atomistic models and X-ray crystallography data,” he says. “This allows the structure determination problem to be formulated as a more principled conformation sampling problem with constraints from the X-ray data as well as other sources such as potential energy.”
“The work Minhuan has done so far took a lot of crystallographic theory and machine learning to realize, and takes big steps in the direction we really want: physical models of how proteins work, parametrized based on experimental data,” Hekstra says.
Molecular dynamics have fascinated Li, who studied colloidal systems and phase transitions as a student at Fudan University in Shanghai. However, analyzing proteins will pose new challenges. “Protein systems have a much smaller length scale and much faster time scale, making the path to particle-level dynamics much more complex,” Li says. “Therefore, we have to spend time thinking about how to approach the data in a more principled and efficient manner.”
When time permits, Li enjoys playing soccer and watching movies to take breaks from working in the lab.
“Minhuan is always curious how things work, whether it is in the physics and biology of proteins, protein design, or the working of the newest machine learning algorithms,” Hekstra adds. “At the same time, he works systematically to get from his questions to mastery of techniques and answering questions. Every time it’s a pleasure to discuss his work. What’s also very nice to see is how much Minhuan has grown into the role of mentor in the lab, working closely with two undergraduate students in the lab, and helping them make rapid progress on quite different projects.”
Li is enthusiastic about continuing his research and grateful to his colleagues. “I would like to extend my gratitude to all Hekstra Lab members and the MCB community,” Li adds. “As an international student, pursuing a PhD during a pandemic has been a difficult experience, but they have made it much easier.”
