The Similarity For Prediction Of Signal Peptides Similarity

Hui Liu,Jie Yang,Lixiu Yao,Yuan Yao

2007-01-01

Abstract:Human assessments of similarity are fundamental to cognition for the world is an orderly enough place that similar objects and events tend to behave similarly. In this paper, we make prediction for identification of the signal peptides by similarity. Most of the former methods for prediciton of signal peptides tended to partition the whole sequences into different segments and extracted features for prediction. They obtained some helpful representations by segments and features, but much useful information was lost and some other problems rose. In this paper, we turn to a new route based on the common assumption, similar things usually behave similarly, we apply the similarity to measure the structure of the signal peptides datasets and more information will be held. Simulataneously, the similarity can assure the high predictive performance by incorporating kernel function and support vector machines (SVM) for the reliable theoretical foundation in statistical learning theory and easily applicable and fast implementations. To meeting with the requirement of SVM, the positive semidefinite is proved by embedding the similarity into Eucliden space. The experiments of prediction for signal peptides also demonstrate that the proposed method is more efficient than conventional methods.