New Feature Vector for Apoptosis Protein Subcellular Localization Prediction

It is widely recognized that the information for determining the final subcellular localization of proteins is found in their amino acid sequences. In this work we present new features extracted from the full length protein sequence to incorporate more biological information. Features are based on the occurrence frequency of di-peptides - traditional, higher order. Naive Bayes classification along with correlation-based feature selection method is proposed to predict the subcellular location of apoptosis protein sequences. Our system makes predictions with an accuracy of 83% using Naive Bayes classification alone and 86% using Naive Bayes classification with correlation-based feature selection. This result shows that the new feature vector is promising, and helps in increasing the prediction accuracy.