Combining Prediction of Protein Aggregation Propensities with Prediction of Other One-Dimensional Properties

Abstract

Protein and peptide aggregation leads to the development of several debilitative disorders, such as Alzheimer's and Parkinson's diseases or pregnancy-related preeclampsia disorder. Misfolded proteins can aggregate forming amyloids that lead to the formation of fibrils and then plaques that disrupt the normal functioning of cells. Because of this the development of accurate methods to predict the aggregation of proteins and peptides is of the utmost importance. There are several known methods for prediction of aggregation propensity of a protein from its sequence, such as AGGRESCAN, FoldAmyloid, FISH, AMYLPRED, and others. Recently we used GOR method originally developed for prediction of protein secondary structure from sequence to predict protein aggregation propensities. We have shown in the past that accurate prediction of protein secondary structure and other one-dimensional structure features is essential for accurate sequence alignment, three-dimensional structure modeling, and function prediction. In the present work we extend this approach by combining prediction of aggregation propensity with predictions of secondary structure, and other one-dimensional properties, such as intrinsically disordered regions, solvent accessible area, and dihedral angles by using a multistep neural-network algorithm. Our preliminary studies show that this method is highly promising and efficient alternative to other known aggregation predicting tools, and leads to improved accuracy of prediction of protein aggregation propensity from sequence.