Recent studies suggest that cardiomyocyte membrane microdomains, caveolae and transverse tubules, play a key role in cardiac arrhythmia. Mutation of caveolin-encoding genes CAV3, co-expressed with genes of caveolae ion channels, leads to a late persistent sodium currents and delayed repolarization stage, called LQT9 disease. A simplified three-current model is created to largely reduce the well-known Pandit rat ventricular myocyte model. The mathematical tractability of the three-current model allows us to conduct asymptotic analysis and efficiently estimate action potential duration. Improvement in the description of the mechanism for caveolae sodium current is incorporated into the three-current model utilizing a probability density approach for the four-state caveolae neck-channel coupling. The prolongation of action potentials and the formation of potential arrhythmia are shown to arise if caveolae neck open probability varies. A minimal model of the Ca2+ spatial distribution of CICR units illustrates the transverse tubule remodeling in failing myocyte causes dysfunction in the Ca2+ profile. With regards to discrimination of protein localization, which is widely used in biological experiments, the bagging pruned decision tree algorithm is tested to be one of the algorithms with best performance on the large data set, and it succeeds in extracting information to be highly predictive on test data. Parallel computation technique is applied to accelerate the speed of implementation in K-nearest neighbor learning algorithms on big data sets.