Mining meaningful and comprehensive molecular characterization of cancers from The Cancer Genome Atlas (TCGA) data has become a bioinformatics bottleneck. Meanwhile, recent progress in cancer analysis shows that multi-omics data can effectively and systematically detect the cancer-related genes at all levels. In this study, we propose an improved particle swarm optimization with dynamic scale-free network, named DSFPSO, to extract features on multi-omics data. The highlights of DSFPSO are taking the dynamic scale-free network as its population structure and diverse velocity updating strategies for fully considering the heterogeneity of particles and their neighbors. Experiments of DSFPSO and its comparison with several state-of-the-art feature extraction approaches are performed on two public data sets from TCGA. Results show that DSFPSO can extract genes associated with cancers effectively.
Keywords: dynamic scale-free network; feature extraction; multi-omics; particle swarm optimization.