A Multi-Instance Multi-Label Learning Approach for Protein Domain Annotation

摘要

Domains act as structural and functional units of proteins, playing an essential role in functional genomics. To investigate the annotation of finite protein domains is of much importance because the functions of a protein can be directly inferred if the functions of its component domains are determined. In this paper, we propose PDAMIML based on a novel multi-instance multi-label learning framework combined with auto-cross covariance transformation and SVM. It can effectively annotate functions for protein domains. We evaluate the performance of PDAMIML using a benchmark of 100 protein domains and 10 high-cycle functional labels. The experiment results reveal that PDAMIML yields significant performance gains when compared to the state-of-the-art approaches. Furthermore, we combine PDAMIML with the other two existing methods by using majority voting, and obtain encouraging results.