Deep‐Learning‐Assisted Stratification of Amyloid Beta Mutants Using Drying Droplet Patterns

Abstract

The development of simple and accurate methods to predict mutations in proteins remains an unsolved challenge in modern biochemistry. It is discovered that critical information about primary and secondary peptide structures can be inferred from the stains left behind by their drying droplets. To analyze the complex stain patterns, deep-learning neuronal networks are challenged with polarized light microscopy images derived from the drying droplet deposits of a range of amyloid beta (1-42) (Aβ42 ) peptides. These peptides differ in a single amino acid residue and represent hereditary mutants of Alzheimer's disease. Stain patterns are not only reproducible but also result in comprehensive stratification of eight amyloid beta (Aβ) variants with predictive accuracies above 99%. Similarly, peptide stains of a range of distinct Aβ42 peptide conformations are identified with accuracies above 99%. The results suggest that a method as simple as drying a droplet of a peptide solution onto a solid surface may serve as an indicator of minute, yet structurally meaningful differences in peptides' primary and secondary structures. Scalable and accurate detection schemes for stratification of conformational and structural protein alterations are critically needed to unravel pathological signatures in many human diseases such as Alzheimer's and Parkinson's disease.