Machine Learning-Assisted Clustering of Nanoparticle-Binding Peptides and Prediction of Their Properties

Bioinspired and biomimetic nanostructures have attracted tremendous interest for theranostic and nanomedicine applications. Among the strategies employed to synthesize these nanostructures, surface functionalization and biomineralization of nanomaterials using peptides stand out due to the wide availability of peptides and their variations as well as the ease of modification process. Effective peptide-based modification of nanomaterials relies on preferential and strong binding between peptides and target nanomaterials. Therefore, the discovery and design of specific peptides with high binding affinity to nanomaterials are essential. Unfortunately, conventional peptide screening methods suffer from shortcomings which render peptide discovery time-consuming, expensive, and cumbersome. Herein, leveraging unsupervised and supervised machine learning, a framework to accelerate peptide analysis is presented. Specifically, more than 1700 nanoparticle-binding peptides are classified into peptide clusters to identify important peptide features to realize higher-affinity binding. In addition, the binding and biomineralization properties of peptides are predicted with high classification accuracy, precision, and recall. This work then proposes the use of unsupervised k-means clustering and supervised k-nearest neighbors algorithms for grouping peptides and predicting their properties, respectively. It is anticipated that the framework originated from this study will further facilitate the rational selection and design of peptides for engineering functional bioinspired and biomimetic nanostructures.