Learning Protein Functions from Bi-relational Graph of Proteins and Function Annotations

Abstract

Abstract. We propose here a multi-label semi-supervised learning algorithm, PfunBG, to predict protein functions, employing a bi-relational graph (BG) of proteins and function annotations. Different from most, if not all, existing methods that only consider the partially labeled proteinprotein interaction (PPI) network, the BG comprises three components, a PPI network, a function class graph induced from function annotations of such proteins, and a bipartite graph induced from function assignments. By referring to proteins and function classes equally as vertices, we exploit network propagation to measure how closely a specific function class is related to a protein of interest. The experiments on a yeast PPI network illustrate its effectiveness and efficiency.

Protein Function Prediction Using Multilabel Ensemble Classification

“…More recently, several semi-supervised multi-label learning methods based on PPI networks were proposed for protein function prediction [15], [16], [18]. MCSL [16] is based on a product graph with an objective function similar to the local and global consistency method [30].…”

“…This augmented matrix often can not be loaded in memory and it's computationally expensive. To address this problem, Jiang [15] proposed a protein function prediction approach called PfunBG, which is based on a birelational graph [22] and label propagation [30]. The birelational graph captures three types of relationships: (i) protein-protein similarities, (ii) function-function similarities, and (iii) protein-function associations, and the association matrix of the bi-relational graph is an (N + C) × (N + C) matrix.…”

“…The inter-function similarity leads to improved prediction accuracy [17], [18] and can be defined in various ways [15], [17], [18]. Here we define the similarity between functions m and n as follows:…”

“…Our results show that the use of a directed bi-relational graph achieves higher accuracy than the undirected one. Our proposed TMC outperforms state-of-the-art protein function prediction approaches; namely, two transductive multi-label classification approaches [15], [18], a transductive classifier [9] and two multi-label multiple kernel learning methods [19], [20]. In addition, the proposed TMEC, which takes advantage of classifier ensembles to make use of multiple heterogeneous data sources, often performs better than TMC.…”

“…This fact should be taken into account in protein function prediction. Several approaches [15]- [18] formulate the protein function prediction problem within a multilabel learning framework. Multi-label learning methods can make use of the dependencies between the different function classes, and they often outperform single-labeled prediction methods.…”