The Budapest Amyloid Predictor and Its Applications.

Area of Science:

• Biochemistry
• Biotechnology
• Neurology
• Structural Biology

Background:

• The amyloid state, characterized by a well-defined β-sheet structure, is crucial in protein aggregation relevant to neurology and biotechnology.
• Current amyloid predictors often use artificial neural networks (ANNs), but their decision-making processes are difficult to interpret.
• Advancements in molecular imaging, such as cryogenic electron microscopy, enhance the understanding of amyloid structures.

Purpose of the Study:

• To develop a highly accurate and interpretable predictor for identifying amyloidogenic hexapeptides.
• To leverage the interpretability of machine learning models to gain biochemical insights into amyloid formation.
• To provide a user-friendly webserver for predicting amyloidogenicity of hexapeptides and their neighbors.

Main Methods:

• Development of a linear Support Vector Machine (SVM) based predictor for hexapeptides.
• Evaluation of predictor correctness, achieving higher than 84% accuracy.
• Implementation of the predictor in the Budapest Amyloid Predictor webserver.

Main Results:

• The linear SVM predictor demonstrates correctness exceeding 84%, matching or surpassing existing ANN-based tools.
• The SVM model's decisions are more interpretable, allowing for the inference of biochemical knowledge related to amyloid formation.
• The Budapest Amyloid Predictor webserver provides predictions for user-input hexapeptides and their distance-1 neighbors.

Conclusions:

• Linear SVMs provide an effective and interpretable alternative to ANNs for predicting protein amyloidogenicity.
• Interpretable models can yield valuable biochemical insights into the sequence determinants of amyloid formation.
• The Budapest Amyloid Predictor offers a practical tool for researchers studying protein aggregation.