A Dynamic-Time Distance Based on Wavelet Decomposition for Subcellular Localization Classification

The use of bioinformatics to predict protein subcellular localization is a significant method to study protein function. In the present study, a novel dynamic-time distance measurement based on wavelet packet decomposition (WPD) was proposed for protein subcellular localization prediction. The protein sequence was firstly converted into multiple numerical signals according to physical/chemical properties. Following signal decomposition into multiple subsignals by wavelet packet decomposition, a comprehensive dynamic-time distance was obtained by the dynamic time warping (DTW) algorithm. Finally, the expected classification can be produced based on the new distance measurement. By introducing DTW, the present algorithm can overcome the shortcoming that traditional methodologies can not measure the similarity of unequal-length sequences. Additional, multi frequency band subsignals can retain more information to achieve accurate results. It was suggested that our algorithm provides superior classification recognition by setting traditional methods as a comparative experiment in the E.coli dataset. It can distinguish cytoplasm and cell membrane proteins that are particularly difficult to be identified by the traditional methodologies.