Data Type Integration for Protein Identification using Kernel Based Classification Methods

The integrated biological data is expected to obtain a higher exactness, better performance and greater robustness compared to single dataset. In this work, we present data integration using kernel-based approach to identify protein class in yeast, ribosomal proteins and membrane proteins. By using intermediate stage of integration, we change the single data source into kernel matrix format. Kernel weighting was used in the establishment of integrated data. We propose three weighting methods approach i.e. KTA (Kernel Target Alignment), FSM (Feature Space-based kernel matrix evaluation Measure), and AI (Alignment Index). We also perform the combination of these three methods. These integrated kernels will be analyzed using Support Vector Machine (SVM). Our proposed data integration methods achieve a higher performance compared to single data source. KTA is the best kernel weighting measurement method and always obtain a better performance to recognize membrane and ribosomal proteins classes than others.