Analysis of four different sets of predictive features for metalloproteins

Metals bound to the protein are important for functional or structural roles. Despite their importance there is a distinct lack of research for identification of metalloproteins from sequence data and their predictive features that help distinguish them from non-metal binding proteins. In this study, four sets of features were analysed in order to see their ability to distinguish between metal and non-metal binding proteins. The analysis was carried out using a novel fuzzy logic method. The results show that the amino acid composition is more capable of distinguishing metal from non-metal binding proteins, than any of the other three features, yielding a predictive accuracy of 69.4%. Cofactors were the least useful feature for distinguishing metalloproteins. However, better results were obtained when physico-chemical and secondary structure features are used, yielding accuracies of 67.8% and 67.1%, respectively. Although the amino acid composition yields the highest predictive accuracy, considering the number of features, the latter two sets of features may be more appropriate for such analysis.