effect
AI system helps researchers piece together one of the largest molecular structures in human cells.
When Pietro Fontana joined the Woo Institute at Harvard Medical School and Boston Children’s Hospital in May 2019, he had one of the world’s most difficult giant jigsaw puzzles in front of him. This involved piecing together a model of the nuclear pore complex, one of the largest molecular machines in human cells.
“It was very difficult from the beginning,” he explains. This complex is called gigantic for a reason. This complex is made up of more than 30 different protein subunits called nucleoporins, with a total of more than 1,000 subunits intertwined in a complex complex.
I think AlphaFold has completely changed the concept of structural biology.
Pietro Fontana Postdoctoral Fellow
So when he first sat down to put AlphaFold to work two years later, along with Alexander Tong of the University of California, Berkeley, who was more familiar with AI systems, he wasn’t sure it would help. But what followed in the summer of 2021 was a somewhat unexpected moment of breakthrough. AlphaFold predicted the previously undetermined structure of nucleoporins, revealing more of the nuclear pore complex in the process. Thanks to AI, they were able to create a nearly complete model of the cytoplasmic loop of the complex.
Model of the nuclear pore complex cytoplasmic loop. Credit: Fontana et al. science 2022
“Many of the components were already well known, but with AlphaFold we also created structurally unknown components,” he says. “I started to realize what a big and useful tool this really was for us. I believe AlphaFold has completely changed the concept of structural biology.”
Molecular scientists like Fontana have devoted decades to deciphering the nuclear pore complex. This is important because it acts as a gatekeeper to everything that goes in and out of the nucleus and is thought to hold the answer to a growing number of serious human diseases, including amyotrophic lateral sclerosis (ALS) and other neurodegenerative diseases. Knowing how the jars are put together could open the door to other groundbreaking discoveries, even life-saving discoveries.
The sheer size of the complex is challenging enough, but the variety of parts adds complexity. “That is one of the biggest challenges in achieving a solution. [clear enough] We can interpret the sequence and structure of the complex,” says Hao Wu, a senior researcher at the institute. Despite the abundance of data, previously only medium-resolution structural images were maintained.
Missing puzzle pieces also hindered progress. Without the full set, it’s hard to tell how the puzzle fits together, Wu says. “To figure out how different protein subunits fit together, we need help with their individual structures,” explains Wu.
This is where AlphaFold changed the game for Wu Lab, which includes Ying Dong and Xiong Pi. By running this on proteins discovered in the eggs of the African clawed frog (Xenopus laevis), which was used as a model system, the team succeeded in mapping the structures of all the different subunits that were hitherto unknown. Wu recalls: “When we started trying, we had no idea whether the predictions would fit the map well,” he said. “But it happened. “It was truly amazing.”
Of course, science is a collaborative effort. Solving a mystery as complex as the nuclear pore complex requires not just teamwork, but the culmination of diligence and persistence from numerous teams around the world. Across the Atlantic, scientists at the Max Planck Institute of Biophysics (MPIBP) and the European Molecular Biology Laboratory (EMBL) in Germany used AlphaFold in conjunction with cryo-electron tomography to model human NPC. What they have achieved so far is a new model that is twice as complete as its predecessor. Now that we have two-thirds of the NPCs, a significant part of the puzzle has been solved, and great progress has been made in understanding how they control what enters and exits the cell nucleus.
A model of the human nuclear pore complex created by scientists at MPIBP and EMBL in Germany. Credit: Agnieszka Obarska-Kosinska
There is still a long way to go. The last third remains. AlphaFold may make solving the rest of the puzzle easier, but scientists are also aware of its limitations. According to Wu, the AI system worked well because the subunits of the nuclear pore complex contain repeating helical structures that are easy to predict. However, for other proteins it may not be that simple.
It is important not to treat AlphaFold or any other AI tool as the be-all and end-all. container. “In fact, AlphaFold can give very strange results,” says Wu. “But if you understand how to predict, you can take that into account. [in the analysis].”
Nonetheless, it is clear that AlphaFold has not only pushed the boundaries of science, but has done so within a timeframe previously thought impossible. “We’re excited that AlphaFold has arrived at the right moment as it significantly speeds up everything we do,” says Fontana.
Fontana P., Dong Y., Pi Science 376, 6598, (2022). DOI:10.1126/science.abm9326.