AI Tool Predicts Toxic Protein Structures Behind Brain Disorders - Neuroscience News

Summary: A new AI tool called RibbonFold is revolutionizing how scientists understand misfolded proteins linked to Alzheimer’s and Parkinson’s. Unlike existing tools such as AlphaFold, which predict correctly folded proteins, RibbonFold is specifically designed to model the twisted, ribbonlike shapes of amyloid fibrils that accumulate in neurodegenerative diseases.

The tool uses physical energy constraints to accurately predict how these toxic proteins evolve into more insoluble and disease-promoting forms. These insights may reshape drug development by allowing researchers to design therapies targeting the most dangerous fibril structures.

Key Facts:

• Specialized AI Modeling: RibbonFold outperforms AlphaFold in predicting misfolded amyloid structures.
• Disease Insight: It reveals how protein fibrils evolve into more insoluble forms over time, possibly explaining late disease onset.
• Drug Potential: Offers a precise tool for targeting harmful protein aggregates in neurodegenerative disorders.

Source: Rice University

A novel artificial intelligence (AI) tool has revealed how disease-linked proteins misfold into harmful structures, a key advance in understanding neurodegenerative disorders such as Alzheimer’s and Parkinson’s.

The study, led by Mingchen Chen of the Changping Laboratory and Rice University’s Peter Wolynes, introduces RibbonFold, a new computational method capable of predicting the structures of amyloids — long, twisted fibers that accumulate in the brains of patients suffering from neurological decline.

The study was published in the Proceedings of the National Academy of Science April 15.

RibbonFold is uniquely tailored to address the complex and variable structures of incorrectly folded proteins rather than functional proteins.

“We’ve shown how AI folding codes can be constrained by incorporating a physical understanding of the energy landscape of amyloid fibrils to predict their structures,” said Wolynes, the D.R. Bullard-Welch Foundation Professor of Science and co-director of the Center for Theoretical Biological Physics.

“RibbonFold outperforms other AI-based prediction tools like AlphaFold, which were trained only to predict correctly folded globular protein structures.”

Eclipsing the gold standard

RibbonFold builds on recent advances in AI-driven protein structure prediction. Unlike tools such as AlphaFold2 or AlphaFold3, which are trained on well-behaved, globular proteins, RibbonFold includes constraints suited to capture the ribbonlike characteristics of amyloid fibrils.

The researchers trained the model using existing structural data on amyloid fibrils then validated it against other known fibril structures deliberately excluded from the training.

Their results demonstrated that RibbonFold outperforms existing AI tools in this specialized domain and reveals previously overlooked nuances in how amyloids form and evolve in the body.

Importantly, it suggests that fibrils may begin in one structural form but may shift into more insoluble configurations over time, contributing to disease progression.

“Misfolded proteins can take on many different structures,” Wolynes said.

“Our method shows that stable polymorphs will likely win out over time by being more insoluble than other forms, explaining the late onset of symptoms. This idea could reshape how researchers approach neurodegenerative disease treatment.”

New frontier in drug development and beyond

RibbonFold’s success in predicting amyloid polymorphs may mark a turning point in how scientists can approach neurodegenerative diseases.

Offering a scalable, accurate method for analyzing the structure of harmful protein aggregates, RibbonFold opens new possibilities for drug development. Pharmaceutical researchers can now target drug design by binding to the most disease-relevant fibril structures with greater precision.

“This work not only explains a long-standing problem but also equips us with the tools to systematically study and intervene in one of life’s most destructive processes,” said Chen, co-corresponding author of the study.

Beyond medicine, these findings offer insights into protein self-assembly, which could impact synthetic biomaterials. In addition, the study resolves a critical mystery in structural biology: why identical proteins can fold into multiple disease-causing forms.

“The ability to predict amyloid polymorphs efficiently may guide future breakthroughs in preventing harmful protein aggregation, a crucial step toward tackling some of the world’s most pressing neurodegenerative challenges,” Wolynes said.

Other authors of this study include co-first authors Liangyue Guo and Qilin Yu along with Di Wang and Xiaoyu Wu of the Changping Laboratory.

Funding: The study had support from the National Science Foundation, the Welch Foundation and the Changping Laboratory.

About this AI and neurology research news

Author: Marcy de Luna
Source: Rice University
Contact: Marcy de Luna – Rice University
Image: The image is credited to Neuroscience News

Original Research: Open access.
“AI tool unlocks long-standing biomedical mystery behind Alzheimer’s, Parkinson’s” by Mingchen Chen et al. PNAS

Abstract

AI tool unlocks long-standing biomedical mystery behind Alzheimer’s, Parkinson’s

The concept that proteins are selected to fold into a well-defined native state has been effectively addressed within the framework of energy landscapes, underpinning the recent successes of structure prediction tools like AlphaFold. The amyloid fold, however, does not represent a unique minimum for a given single sequence.

While the cross-β hydrogen-bonding pattern is common to all amyloids, other aspects of amyloid fiber structures are sensitive not only to the sequence of the aggregating peptides but also to the experimental conditions. This polymorphic nature of amyloid structures challenges structure predictions.

In this paper, we use AI to explore the landscape of possible amyloid protofilament structures composed of a single stack of peptides aligned in a parallel, in-register manner. This perspective enables a practical method for predicting protofilament structures of arbitrary sequences: RibbonFold.

RibbonFold is adapted from AlphaFold2, incorporating parallel in-register constraints within AlphaFold2’s template module, along with an appropriate polymorphism loss function to address the structural diversity of folds.

RibbonFold outperforms AlphaFold2/3 on independent test sets, achieving a mean TM-score of 0.5. RibbonFold proves well-suited to study the polymorphic landscapes of widely studied sequences with documented polymorphisms. The resulting landscapes capture these observed polymorphisms effectively.

We show that while well-known amyloid-forming sequences exhibit a limited number of plausible polymorphs on their “solubility” landscape, randomly shuffled sequences with the same composition appear to be negatively selected in terms of their relative solubility. RibbonFold is a valuable framework for structurally characterizing amyloid polymorphism landscapes.