| Metal ion-binding proteins are special proteins which interact with certain metal ions.Biologists have estimated that about one-third of all known proteins can bind to metal ions.These proteins are quite important in catalysis,regulation and stabilization during various of life processes.Within cells,small changes in the concentration of metal ions may shift their effects from a naturally beneficial function to a harmful one.Therefore,the interaction between metal ions and proteins plays important roles in the normal expression function of proteins.The correct identification of metal ion binding proteins is of great significance for the detection and analysis of pathological features.The traditional biochemical experiments for recognizing of metal ion-binding residues are time-consuming and labor-intensive,which are difficult to satisfy the large-scale applications.As a result,the computation-based methods provide promising solutions for this problem.The main work is as follows:(1)We present a systematic and comprehensive review which relates to the latest progress of the predictions of metal ion and protein interactions.First,we collect the state-of-the-art databases of metal ion-protein interactions.Second,computation-based methods for predicting metal ion binding proteins are analyzed and compared,which are divided into four categories: learning-based,docking based,template-based and meta-learning approaches.The principles,model building methods and application scenarios of each method are discussed in detail.Third,the performance of representative research methods is compared on the benchmark data set.Finally,this paper summarizes the current online prediction tools or offline software packages in the field of metal ion binding proteins.Based on this,we also point out the potential research direction in this field for the end readers.(2)We propose a novel high-throughput method for accurate identification of calcium ion binding residues.Calcium is one of the most abundant elements in human tissues and organs.It is involved in many life activities in human body.We extract several types of features including amino acid composition,evolutionary conservation and solvent accessibility.Besidues that,the incremental feature selection strategy and different machine learning algorithms are used to select the optimal feature subset and classifier on the benchmark training set.We introduce a two-layer model;the first layer is mainly used to accurately predict various types of binding residues;the second layer focuses on reducing the cross prediction error based on the first layer.Next,we compare the accuracy and false positive errors between the proposed method and the current methods on the benchmark test set.In order to further verify the ability of the proposed method to accurately identify calcium ion binding residues,different types of metal ion binding proteins are also tested.We apply the proposed method to the identification and prediction of calcium-bidning residues in the human proteome.The experimental results confirm that the proposed method can effectively recognize calcium-binding residues and provide informative clues to potential new calcium-binding residues.The proposed method has been constructed as a free academic online platform and provides large-scale computing services. |
