Combining Evolutionary Information and Sparse Bayesian Probability Model to Accurately Predict Self-interacting Proteins

Abstract

Self-interacting proteins (SIPs) play a crucial role in investigation of various biochemical developments. In this work, a novel computational method was proposed for accelerating SIPs validation only using protein sequence. Firstly, the protein sequence was represented as Position-Specific Weight Matrix (PSWM) containing protein evolutionary information. Then, we incorporated the Legendre Moment (LM) and Sparse Principal Component Analysis (SPCA) to extract essential and anti-noise evolutionary feature from the PSWM. Finally, we utilized robust Probabilistic Classification Vector Machine (PCVM) classifier to carry out prediction. In the cross-validated experiment, the proposed method exhibits high accuracy performance with 95.54% accuracy on S.erevisiae dataset, which is a significant improvement compared to several competing SIPs predictors. The empirical test reveal that the proposed method can efficiently extracts salient features from protein sequences and accurately predict potential SIPs.