Feature Extraction in Spatially-Conserved Regions and Protein Functional Classification

Abstract

One of the most challenging problems in bioinformatics is prediction of protein functions and structures in unknown protein sequences. The sequence similarity-based approach is the most effective method for the prediction of protein function, but the approach often fails to identify the relevant proteins when similarity does not exist or exists at very low levels. Therefore, it is important to develop prediction and classification methods of protein function without sequence similarity. Our aim is to suggest protein function classification using protein properties without sequence similarity. In this paper, we propose feature extraction in spatially-conserved region sequences and apply high-ranked features through the selection of attributes for the classification of protein function. The experimental results demonstrate that RMSE and MAE rates decrease after low-ranked attributes are discarded from our classification. Our method points out classification using only important short sequences such as motif or conserved regions.