Developing machine-learning-based amyloidogenicity predictors with Cross-Beta DB

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Protein amyloidogenesis is implicated in various diseases and biological functions.
• Accurate computational prediction of amyloidogenicity is crucial for understanding these processes.
• The performance of AI-driven predictors relies heavily on the quality of training data.

Purpose of the Study:

• To create a high-quality database of naturally occurring cross-β amyloids.
• To develop and benchmark machine learning (ML) algorithms for predicting protein amyloid-forming potential.
• To introduce a novel computational tool for assessing amyloidogenicity.

Main Methods:

• Construction of Cross-Beta DB, a curated database of known cross-β amyloids.
• Training and evaluation of multiple ML algorithms using the Cross-Beta DB dataset.
• Development of the Cross-Beta predictor utilizing an Extra Trees ML algorithm.

Main Results:

• The Cross-Beta predictor achieved superior performance compared to existing methods.
• The predictor demonstrated a high F1 score of 0.852 and accuracy of 0.844.
• The developed ML model effectively predicts the amyloid-forming potential of proteins.

Conclusions:

• The Cross-Beta DB database provides a valuable resource for amyloid research.
• The Cross-Beta predictor offers a robust tool for identifying amyloidogenic proteins.
• This advancement may facilitate personalized risk profiling for neurodegenerative diseases and other amyloidoses.