An evolution-based DNA-binding residue predictor using a dynamic query-driven learning scheme

作者：Chai, H.; Zhang, J.; Yang, G.^*; Ma, Z.

来源：Molecular Biosystems, 2016, 12(12): 3643-3650.

DOI：10.1039/c6mb00626d

• PROTEINS
• SITES
• SEQUENCE
• IDENTIFICATION
• MODEL

DNA-binding proteins play a pivotal role in various biological activities. Identification of DNA-binding residues (DBRs) is of great importance for understanding the mechanism of gene regulations and chromatin remodeling. Most traditional computational methods usually construct their predictors on static non-redundant datasets. They excluded many homologous DNA-binding proteins so as to guarantee the generalization capability of their models. However, those ignored samples may potentially provide useful clues when studying protein-DNA interactions, which have not obtained enough attention. In view of this, we propose a novel method, namely DQPred-DBR, to fill the gap of DBR predictions. First, a large-scale extensible sample pool was compiled. Second, evolution-based features in the form of a relative position specific score matrix and covariant evolutionary conservation descriptors were used to encode the feature space. Third, a dynamic query-driven learning scheme was designed to make more use of proteins with known structure and functions. In comparison with a traditional static model, the introduction of dynamic models could obviously improve the prediction performance. Experimental results from the benchmark and independent datasets proved that our DQPred-DBR had promising generalization capability. It was capable of producing decent predictions and outperforms many state-of-the-art methods. For the convenience of academic use, our proposed method was also implemented as a web server at http://59.73.198.144:8080/DQPred-DBR/.

• 出版日期2016
• 单位东北师范大学