Metalloproteins: metal binding predicted on the basis of the amino acid sequence

A protein sequence is often insufficient for knowledge of the chemical formula and the properties of the mature molecule that perform its function. Post-translational modifications are very common and most of them cannot be predicted on the basis of the protein sequence alone. A very common chemical modification of proteins that is not directly encoded by a single gene is the complexation with metal cations. Here it is shown that the uptake of metal ions (calcium, cobalt, copper, iron, magnesium, manganese, nickel or zinc) by proteins can be predicted on the basis of the amino acid composition, by using a mixture of several simplified amino acid alphabets and by employing machine learning methods, with 70-90% accuracy, depending on the type of metal. Not only is it possible to predict if a protein requires a certain metal ion but it is also possible to discriminate amongst various metal species. These results are likely to be useful in structural proteomics, by improving the experiment success rate, and in comparative genomics, where it is interesting to compare metal-ion contents in different organisms. It is particularly important that these predictions can be made when homology-based annotations are impossible.