Researchers Developed AI Tool to Identify Amyloid Cores
The new model identifies structural amyloid-core regions in proteins with greater accuracy than previous methods.
Updated on Oct. 1, 2026 in Artificial Intelligence
Researchers have developed AmyloCore-ML, a machine learning predictor designed to identify structural amyloid-core regions within proteins. This research-stage development relies on protein language model embeddings and has been benchmarked against existing predictors.
Why it matters
The development addresses a critical limitation where prior predictors failed to accurately map residues within disease-associated fibril structures. This tool provides a more precise method for identifying these structural regions, which are essential to understanding protein aggregation processes.
AmyloCore-ML achieved an AUROC of approximately 0.88 and an AUPRC of 0.85, outperforming the ESM-2/ExtraTrees W21 model which recorded 0.833 AUROC and 0.751 AUPRC. The model shows a mean peak distance of 12.4 residues, calculated based on experimentally determined fibril structures.
The details
The model uses protein language model embeddings—vector representations of protein sequences that capture structural and evolutionary information—to analyze residue interactions. Researchers trained the system using experimental data from the Amyloid Atlas, designating ordered residues as core regions and unresolved residues as non-core. This allows the model to map the precise structural architecture of amyloid fibrils, a type of protein aggregate linked to various diseases.
Timeline
The findings were published on October 1, 2026.
The Tech Race
This development follows a pattern set by the Amyloid Atlas, which serves as a foundational dataset for mapping protein aggregation. AmyloCore-ML marks an advancement over established predictors like CrossBeta and AggrescanAI by improving the accuracy of structural identification.
Researchers and developers can access the model now via an interactive Google Colab notebook for protein analysis. It is currently available as a research tool rather than a clinical diagnostic product.
The takeaway
This model provides a more robust way to predict amyloid-core regions, closing a gap in current structural biology workflows. Interested researchers should monitor the model's application to specific disease-associated proteins to see how its predictions correlate with further experimental data.
Further reading
For broader context on how machine learning is mapping biological structures, explore the latest research in Artificial Intelligence.
