Predicting Signal Peptides and Their Cleavage Sites Using Support Vector Machines and Improved Position Weight Matrixes

In this paper, we develop a method for predicting signal peptides and their cleavage sites. Unlike other published work, we divide proteins into two segments and calculate the amino acid compositions on both segments. After that, we hybridize the pseudo amino acid compositions (PseAAs) to the feature vectors. Using support vector machines (SVMs) to train the datasets, we get better results than those with the optimized evidence-theoretic K nearest neighbor (OET-KNN) classifier. The overall rate of correct prediction for signal peptides is over 97%. For identifying cleavage sites, we use the scaled window proposed by Chou to extract cleavable secretory segments and non-cleavable secretory segments and improve the position weight matrix (PWM) method proposed by Hiller et al.. By hybridizing the scaled window and PWM methods, the correct prediction for signal peptides cleavage sites is also better or comparable to other methods.