Predicting Protein-protein Interactions from Protein Sequences Using Probabilistic Neural Network and Feature Combination

Identifying Protein-protein Interactions (PPIs) can provide a deep insight in cellular processes and biochemical events. Although many computational methods have been proposed for this work, there are still many difficulties due to the high computation complexity and noisy data. In this paper, a novel method based on Probabilistic Neural Network (PNN) with feature combination was proposed for PPIs prediction. PNN is a statistic model and is robust to noise. It need not to be trained compared with other computational models such as Artificial Neural Network (ANN) and Support Vector Machine (SVM). So it is very fast and can deal with large scale noisy PPIs data more properly. In addition, in order to obtain the more informative features from protein pairs, three most import physicochemical properties were adopted for featuring, then the three different features are combined as the input for PNN training and the different combinations were tested to get the best combination. Experiments show that our proposed method produces the best performance compared with the other popular methods.