A New Method For Predicting Disulfide Bonds Based On Protein Sequence Information

Protein is the material basis of all life activities,and it plays an important role in life phenomenon and physiological function.Proteomics is an important branch belonging to bioinformatics.By researching of this branch,humans can understand the nature of life process better,and can gain the theoretical foundation of the prevention and treatment of diseases,and as a result,our living quality can also be improved.In recent years,governments and researchers from all over the world attaches great importance to this subject.With the coming of postgenomics,Proteomics has become the main research direction of bioinformatics.Protein consists of huge of Amino acid residues via complex chemical reactions,containing complex structures and functions,which functions are realized by thousands of combinations with interactions among 20 different amino acids.Disulfide bond plays an important role in maintaining the constructs and functions of protein,whose bonding state and connectivity pattern deserve extensive research.In recent years,with the rapid development of protein sequencing technology,the number of known protein sequences as well as the disulfide bonds is growing explosively.Using traditional methods based on Biology testing disulfide bond not only costs much,but also spends much time.So,predicting disulfide bond accurately and fast by coming up intelligent algorithms is the priority.In recent years,different kinds of algorithms based on Machine Learning have been used on disulfide bond prediction.Cysteine pairs in a protein sequence have two states(they can form a disulfide bond or not),which belongs to a typical binary classification problem.The former researcher experimentally addressed the problem of disulfide bond prediction through feature extraction and feature selection,and then gain final results through classifier.When researching the connectivity patterns,this paper divided the features into amino acid residues features,cysteine features and global features to find the most effective feature group.In the process of feature selection,firstly,calculating each feature's score(including variance score,Laplacian score and Fisher score)to choose representative features,then carrying out correlation feature selection.Finally,predicting connectivity patterns through different classifiers.Comparing with former research,this paper carried out correlation feature selection.The algorithm and benchmark datasets are freely available at:https://github.com/snep5021/Disulfidesprediction.