A Novel Neighborhood Model to Predict Protein Function from Protein- Protein Interaction Data

Proteins are the second largest portion of a cell after water, playing a wide variety of key roles in the cellular activities. Due to low-efficiency in determining protein functions, however, accurately and cost-effectively identifying functions of a protein is still one of major challenge in the post-genomics era. In an attempt to close this gap, we presented a novel neighborhood model to predict protein functions from protein-protein interaction data. The new method takes into consideration functions of directly interacting proteins as well as ones of indirectly interacting proteins, and also weighted interactions. The jackknife test on 4662 proteins in the S.cerevisiae shows that our method outperforms other two neighborhood models: the neighbor counting method and Chi-square method. The experimental results also show that the functional information of indirectly interacting proteins are greatly decayed to infer protein function, and the function categories involving interactions are predicted with higher accuracy than those not involving interactions.