This article presents an approach in bioinformatics data analysis and exploration that improves classification accuracy by learning the inner structure of the data. The diseases studied in bioinformatics (diagnostic, prognostic etc. studies) often have the known or yet undiscovered subtypes that can be used while solving bioinformatics tasks providing more information and knowledge. This study deals with the problem above by studying inner class structures (probable disease subtypes) using a cluster analysis to find classification subclasses and applying it in classification tasks. The study also analyses possible cluster merges that would best describe classes. Evaluation is carried out using four classification methods that can be successfully used in bioinformatics: Naïve Bayes classifiers, C4.5, Random Forests and Support Vector Machines.