Prediction of Golgi-resident protein types by using feature selection technique

• A novel computational method was proposed for predicting the types of Golgi-resident proteins.
• The overall accuracy achieves 85.4% by using optimized 2-gap dipeptides.
• An effective tool, subGolgi, was constructed for predicting Golgi-resident proteins and their types.

The functions of Golgi apparatus are to store, package and distribute proteins. Knowing the type of a Golgi-resident protein will provide in-depth insight into its function. In this study, we developed a support vector machine-based method to identify the types of Golgi-resident proteins by using only amino acid sequence information. A strictly and objective dataset including 137 proteins with the sequence identity < 25% was used for training and testing the support vector machine. The analysis of variance was proposed to find out the optimized feature set. In the leave-one-out cross-validation, the maximum overall accuracy of 85.4% was achieved with the area under the receiver operating characteristic curves of 0.878. The results demonstrate that the proposed method can be used to discriminate the types of Golgi-resident proteins. An on-line server subGolgi is freely available at http://lin.uestc.edu.cn/server/subGolgi2.