Identifying The Types Of Ion Channel-Targeted Conotoxins Based On Feature Selection Techniques

One important nosogenesis of several diseases is the dysfunction of the ion channel. Thus, the ion channels are regarded as important targets in the treatment of numerous diseases. As one of the marine toxins which have specificity and affinity towards ion channels, conotoxins have been one of the most popular research hotspots and are called pharmaceutical treasure in the ocean.Conotoxin, a kind of neurotoxin, has disulfide-rich small peptide with high diversity of amino acid composition. Because the most of them have high specificity and affinity towards ion channels, conotoxins exhibit the tremendous potential in the development of neurological drugs. Although there are the plenty of conotoxins in cone snails, the published literatures about conotoxins were very few. Especially, few bioinformatics studies were focused on these important toxins. With the rapid rise of proteomics, it is more and more important to identify a new sequenced conotoxins by using bioinformatics methods. As an important tool to obtain and process information, machine learning approaches play an important role in bioinformatics. Currently, researchers have proposed some methods to predict the superfamilies of conotoxins based on amino acid sequences. However, these methods can not directly predict the ion channel activities of conotoxins. Based on this consideration, this thesis aims to develop an efficient bioinformatics method to recognize different ion channel-targeted conotoxins.In this thesis, we focused on the conotoxin sequences and their function. Based on the theory of bioinformatics and machine learning approach, the feature selection techniques were used to extract the optimal parameters for discirminating K+, Na+and Ca2+channel-targeted conotoxins. And we evaluated the performance of different models and compared the results of different algorithms.Firstly, we constructed reliable benchmark datasets by searching experimental-confirmed conotoxins. Subsequently, based on dipeptide compositions, the binomial distribution and F-score were used to filter out the best feature set. By inputing these parameters into radial basis function network and support vector machine, the types of ion-channel targeted conotoxins can be predicted. Results show that both methods can achieve encouraged results. Especially, in jackknife cross-validation, the model which based on SVM and F-score can correctly predict over90%conotoxins. Based on this model, a free online web-server called ICTCPred was constructed and will offer experimental scientists a convenient and reliable tool to predict ion channel targeted conotoxins. This server can also lower down the cost of relevant researches and provide theory basis for relevant researches.