Predict Protein Function Based On Frequent Functional Pattern

The21st century is the era of life science and information science, bioinformatics，as across and emerging discipline, came into being in such an environment, and is the constantdevelopment. Generally speaking, bioinformatics is the study of the acquisition of thebiological information, computing, storage, analysing and interpretation of a discipline. It isthrough the comprehensive utilization of biology, computer science and IT, to explain thelarge number and complexity biological data generated by biological problems. The analysisof protein interactions is the basis for the understanding of cells and tissues, biologicalprocesses and protein function. According to the experiments, proteins rarely work alone, butseveral proteins involved in the same cellular processes and the interaction function. We canpredict the non-annotated protein’ function by the annotated interacting proteins. The researchof the protein-protein interaction network can provide the insight on the predicting of thefunction of proteins.Predicting protein function from protein interaction networks has been challengingbecause of the complexity of functional relationships among proteins. Most previous functionprediction methods depend on the neighborhood of or the connected paths to known proteins.However, their accuracy has been limited due to the functional inconsistency of interactingproteins. In this paper, we propose a novel approach for function prediction by identifyingfrequent patterns of functional associations in a protein interaction network. A set of functionsthat a protein performs is assigned into the corresponding node as a label. A functionalassociation pattern is then represented as a labeled subgraph. Our frequent labeled subgraphmining algorithm efficiently searches the functional association patterns that occur frequentlyin the network. It iteratively increases the size of frequent patterns by one node at a time byselective joining, and simplifies the network by a priori pruning. The function prediction isperformed by matching the subgraph, including the unknown protein, with the frequentpatterns analogous to it. By leave-one-out cross validation, we show that our approach hasbetter performance than previous methods in terms of prediction accuracy. The frequentfunctional association patterns generated in this study might become the foundations ofadvanced analysis for functional behaviors of proteins in a system level. In this paper, we explore the efficient discovery of frequent patterns of functionalassociations in a protein interaction network for the purpose of function prediction. A set offunctions that a protein performs is assigned to the corresponding node as a label. Afunctional association pattern is then represented as a labeled subgraph of the network. Underthe assumption of downward closure of frequency, i.e., a functional association pattern p isfrequent if and only if p and all possible subpatterns of p are frequent, the a priori algorithmcan efficiently find frequent functional association patterns. We present a frequent labeledsubgraph mining approach, derived from the a priori algorithm and upgraded for theapplication to complex interaction networks. In a previous work, the a priori algorithm hasbeen used for finding subgraphs that occur frequently over a set of graphs. However, it isdifferent from our subject because it examines how many graphs in a graph database containthe subgraph of interest, whereas we investigate how many times the subgraph occurs in asingle graph. Our algorithm is composed of two major processes: selective joining foriterative increments of the pattern size and a priori pruning infrequent patterns.