(675f) Multiscale Prediction of Aggregation and Solubility of Amyloid-Derived Peptides

The design of proteins and peptides for therapeutic or industrial use has become increasingly important. However, design of protein/peptide-based products is complicated by competing factors affecting their efficacy and usability, such as solubility and stability. For this reason, accurate methods for the prediction of peptide properties (e.g., solubility) based on amino acid composition alone, are highly desirable. In particular, amyloidogenic sequences are of significant concern for peptide solubility due to their role in the formation of fibrous aggregates. In this study, we have developed multiscale models aimed at identifying factors that contribute to aggregation and overall solubility of amyloid-derived peptides. We have utilized amyloid peptide aggregation datasets from the literature, as well as, developed our own dataset with more quantitative solubility measurements. We produced over 40 suspected amyloid peptides containing between 4 and 11 amino acids and measured the solubility of each peptide using a modified version of the LYophilisation Solubility Assay (LYSA). To develop models of potential aggregate structures we applied a novel flexible-docking algorithm, based on black-box optimization. Peptide aggregate models were built hierarchically, where initial models started with individual peptide molecules and then used to build models of aggregates of increasing size. Structure based features, including intermolecular interactions, solvation free energy, and solvent accessible surface area, were extracted using a combination of statistics-based force fields and high-fidelity biophysical calculations. SVM regression and classification was used to connect structure-based features to experimental aggregation and solubility observations, and non-linear feature selection was applied to identify the structural features that are the most predictive of peptide aggregation and solubility. Comparison and benchmarking with existing models will also be discussed.