Bioinfo Chem

Abstract

Metal ions are essential for metalloproteins to perform their catalytic or structural functions. To understand their role in protein function, it is important to identify metal ion-binding sites. Because experimental identification is labor-intensive and time-consuming, computational methods are expected to be used in the prediction of protein–metal ion-binding sites. A range of computational methods have been proposed to predict metal ion-binding sites from protein sequences. In this study, we implemented two methods of predicting metal ion-binding sites for Ca²⁺, Co²⁺, Cu²⁺, Cu⁺, Fe³⁺, Fe²⁺, Hg²⁺, Mg²⁺, Mn²⁺, Ni²⁺, and Zn²⁺ from amino acid sequences. One is a homology-based method, and the other is a machine-learning method. The homology-based method predicts the binding sites from homologous sequences obtained by a protein–protein basic local alignment search tool (BLASTP) search. The machine-learning method uses a support vector machine with three protein sequence features. Our results showed that the homology-based method achieved an accuracy of 0.9905 and a specificity of 0.9978, while the machine-learning method showed balanced performance with regard to accuracy, sensitivity, and specificity. Especially, the sensitivity of the machine-learning method was 0.8239, and many metal ion-binding sites were predicted only by the machine-learning method.

Keywords: protein, metal ion, binding site prediction, machine learning, homology search