Massive experimental quantification of amyloid nucleation allows interpretable deep learning of protein aggregation

Area of Science:

• Biochemistry
• Computational Biology
• Genetics

Background:

• Protein aggregation is implicated in over fifty human diseases.
• Existing methods for predicting protein aggregation are limited by small, biased datasets.
• Accurate prediction of protein aggregation is crucial for both understanding disease and biotechnology applications.

Purpose of the Study:

• To address the data shortage in protein aggregation prediction.
• To develop an accurate computational model for predicting amyloid nucleation from protein sequence.
• To provide an interpretable deep learning model for analyzing protein aggregation propensities.

Main Methods:

• Experimentally quantified amyloid nucleation propensity for over 100,000 random protein sequences.
• Trained a novel convolution-attention hybrid neural network (CANYA) on the generated dataset.
• Utilized genomic neural network interpretability techniques to analyze the model's decision-making process.

Main Results:

• The large-scale dataset revealed limitations in existing protein aggregation prediction methods.
• The developed CANYA model accurately predicts amyloid nucleation from protein sequence.
• Interpretability analysis provided insights into the learned sequence grammar governing amyloid nucleation.

Conclusions:

• Massive experimental analysis of random sequence spaces is a powerful approach for biological discovery.
• CANYA offers an interpretable and robust computational tool for predicting protein amyloid nucleation.
• The findings advance our ability to understand and engineer proteins, with implications for disease and biotechnology.