Alzheimer’s disease — the most common cause of dementia — includes the accumulation of amyloid fibers. Molecular modeling efforts have focused on identifying critical sequence motifs in amyloid that regulate aggregation to guide the discovery of natural or synthetic inhibitors. Deep Mutational Scanning (DMS) is a high-throughput method to model changes at every position in the amino acid sequence, but it has limitations.
Eric Gamazon, PhD, and colleagues have now applied deep learning, a machine learning method based on neural networks, to model the mutational effect of a pathogenic amyloid beta peptide on aggregation-related traits. They trained an array of neural network architectures on DMS measurements and found that Graph Convolutional Neural Networks improved identification of known disease-causing mutations relative to the original DMS analysis.
The study, reported in Computational and Structural Biotechnology Journal, suggests that neural networks will be useful for providing direct support for protein engineering or genome editing and to facilitate therapeutic design.
Co-authors of the study included Bo Wang, PhD, and Shahab Razavi, PhD. The research was supported by grants from the National Institutes of Health (HG010718, HG011138, AG068026, GM140287). Gamazon is a Life Member of Clare Hall, University of Cambridge.
