The World Health Organization (WHO) reported that over 8.5 million people globally had Parkinson’s disease (PD) as of 2019, with 5.8 million disability-adjusted life years (DALYs) recorded. This represents an 81% increase since 2000. PD-related deaths also rose by more than 100% to 329,000 in the same period.
Researchers at the Agency for Science, Technology and Research (A*STAR) are using AlphaFold, an AI program developed by DeepMind, to investigate how autoantibodies interact with the STIP1 protein. DeepMind, a subsidiary and a major research arm of Alphabet, developed AlphaFold to predict protein structures, solving a long-standing scientific challenge.
“Parkinson’s disease is a neurodegenerative disorder that affects physical mobility, but early diagnosis remains challenging,” said Jackwee Lim, a researcher at A*STAR, during a virtual media briefing. “STIP1, a stress-induced phosphoprotein, is widely expressed in different tissues, particularly in the brain and kidney.”
As a co-chaperone protein, STIP1 plays a crucial role in proper protein folding, ensuring the correct structure of other proteins. Inspired by a journal article, A*STAR began using AlphaFold around three to four years ago and has since incorporated it into ongoing research.
Understanding protein structure prediction
“Proteins are molecular machines in the body that perform almost all functions,” explained Dhavanthi Hariharan, product manager, AlphaFold. “Each protein is made of a chain of amino acid units that form a specific 3D structure. This 3D structure is crucial for the protein’s function.”
According to Hariharan, traditionally, determining a protein’s 3D structure was extremely time-consuming an may take even 20 years to understand hemoglobin’s structure. It also required extensive experimental methods and involved years of research.
AlphaFold’s breakthrough was the hypothesis that: “If you know the sequence of amino acid units in a protein, that alone should be enough to predict its structure.”
The key innovation is that AlphaFold can now take an amino acid sequence as input, predict an accurate, high-quality protein structure, produce results in minutes instead of years, and provide structural insights that previously required billions of research hours.
“In essence, protein structure prediction is like figuring out how a complex molecular puzzle fits together, using only the list of puzzle pieces, instead of physically assembling it through lengthy experimental processes,” Hariharan said.
Accelerating protein research
AlphaFold uses deep learning to predict protein structures from amino acid sequences. It has significantly accelerated protein research by making structural data more accessible. The AlphaFold Protein Structure Database, maintained by DeepMind, now hosts over 200 million protein structures, including nearly all known to science. The data is freely available, which benefits researchers globally.
David Baker, Demis Hassabis, and John Jumper received the Nobel Prize for Chemistry in 2024 for their pioneering work on AlphaFold. The technology’s capability to predict protein folding has transformed the field, enabling new insights into molecular biology.
Visualizing protein interactions
A*STAR created a 3D model of the STIP1 protein, illustrating its different domains. The researchers also mapped the binding domains of related proteins HSP70 and HSP90.
“We are able to understand how autoantibodies interact with the STIP1 protein, visualizing how these interactions might disrupt normal protein folding and potentially contribute to protein aggregation in Parkinson’s disease,” said Lim.
The ultimate aim is to leverage these findings to develop blood diagnostics capable of measuring STIP1 autoantibodies in Parkinson’s disease patients. Lim noted that AlphaFold’s insights into protein interactions and structures could lead to a more comprehensive understanding of the disease.
Extending AlphaFold’s impact
The AlphaFold database is widely used worldwide, including in the Philippines, where over 7,900 users utilize it for various research purposes. One example is scientists at the University of the Philippines Manila use AlphaFold to predict protein structures relevant to vaccine design and evaluation against deadly animal viruses. At the International Rice Research Institute, scientists utilize AlphaFold predictions to visualize protein structures and help interpret phosphorylation sites in rice, which is crucial for understanding genetic expression and improving crop resilience.
Addressing ethical concerns
As with any advanced technology, AlphaFold raises ethical considerations. Hariharan explained that the team at DeepMind conducted thorough expert reviews to evaluate potential risks before releasing the tool. They ensured the technology would be freely accessible, especially for underfunded sectors, and made the interface user-friendly for non-technical researchers.
“Creating something harmful using just protein structures is extremely difficult,” Hariharan said. “It requires additional tools, resources, and specialized knowledge. More importantly, it significantly restricts the likelihood of malicious use.”
To enhance accessibility, AlphaFold’s interface features color-coded confidence metrics that are color-blind friendly. The platform also includes educational resources to make the complex technology more approachable.
“It was aimed to make complex technology accessible to non-technical researchers,” Hariharan said. “We carefully weighed potential risks against potential benefits and concluded that the technology’s positive applications in fields like medicine, agriculture, and scientific research far outweigh potential misuse risks.”
Future directions
Following AlphaFold’s success, the next major challenges in computational biology revolve around protein function prediction and genomics research. Moving beyond structural mapping to understanding the functions of proteins is seen as a key step forward.
Researchers are also exploring how AI techniques could expand to encompass genomics, aiming to understand how DNA works and how genetic mechanisms interact. This transition from structural prediction to functional analysis marks a significant shift in the field.
The focus is shifting from purely structural predictions to functional interpretations, aiming to develop a deeper understanding of how proteins behave in complex biological systems. This approach could ultimately lead to breakthroughs not only in treating neurodegenerative disorders like Parkinson’s but also in agriculture and genetic research.
By leveraging AI-driven tools like AlphaFold, researchers are not only making strides in protein research but also setting the foundation for broader applications in biology, paving the way for innovations in diagnosing and managing diseases.