the PSN (Pocket Similarity Networks) is a software to describe the similarity among these pockets with different thresholds, which are the measurements of the similarity. In order to explore the relationship between the pocket similarity and the GO terms similarity, we statistics the percentage of proteins with similar pockets corresponding to identical GO terms and semantic similarity of the GO functions. Then, the topology structure of the network is used to predict functions. The closest neighbors of pocket and the connected components of the networks would be used to predict a target protein’s functions (GO terms) by a scoring scheme. We test the performance of the method by cross-validated experiments on large scale proteins come from different families. We also predicted the GO functions of some un-annotated proteins.
:: MORE INFORMATION
Zhi-Ping Liu, Ling-Yun Wu, Yong Wang, Luonan Chen, and Xiang-Sun Zhang.
Predicting gene ontology functions from protein’s regional surface structures.
BMC Bioinformatics, Vol. 8, 475, 2007