Predictive Analytics for Biomineralization Peptide Binding Affinity

Abstract

The rational design of biomineralization peptides for the synthesis of inorganic nanomaterials remains a challenging endeavor in biomimetics. The difficulty arises from the multiple factors that influence the affinity of the peptide towards a particular surface. This study presents classification and regression models of biomineralization peptide binding affinity for a gold surface using support vector machine. It was found that the Kidera factors, in particular those related to the extended structure preference, partial specific volume, flat extended preference, and pK-C of the peptide, are important descriptors to predict biomineralization peptide binding affinity. The classification model exhibited an overall prediction accuracy of 90% and 83% for the regression model. This highlights the reliability and accuracy of the formulated models, while requiring a reasonable number of descriptors. The created predictive models are steps in the right direction towards the further development of rational biomineralization peptide design.