Prediction of G-protein-coupled receptor classes with pseudo amino acid composition / 生物医学工程学杂志

G-protein-coupled receptors (GPCRs), the largest family of cell surface receptors, play an important role in the production of therapeutic drugs. The functions of GPCRs are closely related to their classification and subclassification. It is difficult to obtain the spatial structure of GPCRs sequence by experimental approaches. It is highly desired to develop powerful tools and effective algorithms for classifying the family of GPCRs. In this study, based on the concept of pseudo amino acid composition (PseAA) originally introduced by Chou, approximate entropy (ApEn) of protein sequence as an additional characteristic is used to construct PseAA. A 21-D (dimensional) PseAA is formulated to represent the sample of a protein. Fuzzy K nearest neighbors (FKNN) classifier is adopted as prediction engine. The datasets in low homology are used to validate the performance of the proposed method. Compared with prior works, the successful rates achieved of our research are the highest. The test results indicate that the novel approach can play the role of a compliment to many of the existing methods, which promises to be a useful tool for GPCRs function prediction.

Protein secondary structure prediction based on maximum entropy model / 生物医学工程学杂志

Secondary structure prediction plays an important role in function prediction of protein. In this paper, maximum entropy model is used to predict protein secondary structure. We build feature function sets based on the influential factors which are crucial to the states of secondary structure of residues in protein sequence. Multi-factors are taken into account in the model, including charge of amino acids, conformational parameter for the states of secondary structure, short and long ranges of interaction of residues in sequence. As such, multi-source information is integrated into a single probability model by the method. Compared with the reported methods, our method gets a higher accuracy rate in predicting protein secondary structure. The results demonstrate that the proposed method is practical.

Protein structural class prediction with binary tree-based support vector machines / 生物医学工程学杂志

A new mutil-classification method based on binary tree SVM (BT-SVM) is presented to predict protein structural class. The protein sequence, which is represented by 26-D vector, is used as input vector. BT-SVM method resolves unclassifiable regions for multiclass problems which can not be solved by SVM. Self-consistency and cross validation test are used to verify the performance of the proposal method on two benchmark datasets. Satisfactory test results demonstrate that the new method is promising. The Jackknife results of the new method are compared with the existing results on the same datasets. The results of the new method are almost the same as the ones of the best exiting method. It illuminates that the new method has good prediction performance and it will become a useful tool in protein structure class prediction.