Classification of apoptosis proteins by discriminant analysis

Abstract

Objective: Prediction of the locations of apoptosis proteins which are divided into four locations as cytoplasmic proteins, plasma membrane-bound proteins, mitochondrial inner-outer proteins and other proteins; that have different biological function in each location. Methods: In this paper, we have encoded protein sequences as amino acid composition, thereby protein sequences are expressed via high dimensional structure, and consequently this case causes computational complexity. Principal component analysis has been used to reduce the dimension of apoptosis protein sequences. After this preprocessing step, apoptosis proteins are classified by linear discriminant analysis and fuzzy linear discriminant analysis. Results: The overall prediction accuracies for linear discriminant analysis and fuzzy linear discriminant analysis are obtained as 16.3% and 80.2%, respectively. Apoptosis proteins assigned as testing data set represent overlapping data structure, therefore linear discriminant analysis yields unsuccessful result. Since fuzzy logic is appropriate for classification and clustering of overlapping data, fuzzy linear discriminant analysis, the integration of fuzzy logic and linear discriminant analysis, gives satisfying prediction accuracy rates. Conclusion: It can be argued that the association of fuzzy approach with other classical methods can yield higher and more robust prediction accuracy rates for the classification problems of apoptosis proteins.