| With the progress of genome projects and other high throughput applications, biological data have been growing exponentially. Consequently, data management and data mining have become indispensable components of biology. Computational analyses are being widely used in not only large genomics and proteomics projects, but also in studies of individual protein families. In this project, I took an "in silico" approach to collect, manage and explore the structural and functional data of voltage-gated potassium channels (VKCs). VKCs sense change in transmembrane voltage and open to allow potassium ions to pass through an ion-selective pore. They play critical roles in electrically excitable cells. Dysfunctional VKCs are related to diseases including epilepsy and cardiac arrhythmia. I first collected biological information on available VKCs from GenBank, Swissprot, and journal articles. These data and related analyses were stored in a relational database, called the voltage-gated potassium channel database (VKCDB http://vkcdb.biology.ualberta.ca ). Using the collected data, I then built a predictor using a k-nearest neighbor classifier and feature selection techniques. The predictor successfully predicts the voltage sensitivity of a VKC based on its amino acid sequence with a mean absolute error of 7.0mV, and has been validated by permutation tests and independent experimental data. During the learning process, a number of residues were identified as being critical structural elements for modulating voltage sensitivity of VKCs. The methods I used in constructing VKCDB and building the computational model are not specifically tailored for VKCs; they can be easily generalized for study of other protein families. |
