Protein folds recognized by an intelligent predictor based‐on evolutionary and structural information. Issue 4 (27th October 2015)

• Record Type: Journal Article
• Authors: Cheung, Ngaam J.; Ding, Xue‐Ming; Shen, Hong‐Bin
• Is Part Of: Journal of computational chemistry. Volume 37:Issue 4(2016)
• Page Start: 426
• Page End: 436
• Publication Date: 2015-10-27
• Subjects: protein fold recognition -- particle swarm optimization -- neural network -- heterogeneous search -- NiRecor; Chemistry -- Data processing -- Periodicals; 542.85
• DOI: 10.1002/jcc.24232
• Languages: English
• ISSNs: 0192-8651

Abstract

Abstract : Protein fold recognition is an important and essential step in determining tertiary structure of a protein in biological science. In this study, a model termed NiRecor is developed for recognizing protein folds based on artificial neural networks incorporated in an adaptive heterogeneous particle swarm optimizer. The main contribution of NiRecor is that it is a data‐driven and highly‐performing predictor without manually tuning control parameters for different data sets. In biological science, since evolutionary‐ and structure‐based information of amino acid sequences is greatly important in determination of tertiary structure of a protein, accordingly, in NiRecor we employ two different feature sets, which involve position specific scoring matrix and secondary structure prediction matrix, to predict the structural classes of protein folds. The experimental results demonstrate the proposed method is powerful in predicting protein folds with higher precisions by improvements of 1.1 ∼7.8 percentages on three benchmark datasets by comparing with several existing predictors. © 2015 Wiley Periodicals, Inc. Abstract : As an outstanding issue in protein science, protein‐fold recognition is highly important for determining the tertiary structure of a protein from its primary sequence. How does one correctly recognize the protein folds from sequences? A swarm‐optimized predictor—a data‐driven predicting method—is proposed to solve this scientific problem. Without manually tuning parameters, it avoids laborious work on finding an appropriate predictor for the problem and exhibits a good performance, even on different protein datasets.