Detecting DNA-binding domain from sequence and secondary structure Information Using Kernel-based Technique

DNA-binding proteins play an important role in various intra- and extra-cellular activities. The key in the protein is DNA-binding region also called DNA-binding domain (DBD). However, it is hard to search the DBDs by means of homology search or hidden Markov models because of a wide variety of the sequences. In this work, we develop a kernel-based machine learning method by combination of multiple ¿1-vs-1¿ binary classifiers for DNA binding domain prediction. Our result shows that 93.73% accuracy is achieved for multicategory classifier and no less than 90% accuracy for each binary classifier. By comparison, our classifier performs better than other machine learning methods.