Modeling Protein Functional Properties from Amino Acid Composition

Abstract

Physicochemical and functional properties of proteins were modeled as a function of the contributions of each of the 20 coded amino acids to three (z-scores) or five (extended z-scores) amino acid principal properties using partial least squares regression. The five term models were in all cases stronger in both fit and prediction than the three term models, indicating that useful information is contained in the fourth and fifth property scores. Models predicting protein hydrophobicity (R = 0.932), viscosity (R = 0.737), and foam capacity (R = 0.880) from amino acid composition rather than sequence were obtained. It is likely that additional functional and physicochemical properties of proteins can be modeled in this way.