Biomedical Data Analysis Based on Multi-view Intact Space Learning with Geodesic Similarity Preserving

Recent development of technologies has made it easy to obtain a large amount of multi-view biomedical data from different sources or different preprocessing methods, such as the genome-wide data or electroencephalography (EEG) data. Multi-view intact space learning (MISL), which integrates the complementary information from multiple views to discover a latent intact representation of the data, has shown to be effective in computer vision and data mining tasks. However, it fails in biomedical data analysis since it loses geometric information when projecting data from the original space to the intact space. To overcome this problem, we propose a multi-view intact space learning with geodesic similarity preserving (MISL-GSP). The method first builds one KNN graph for each view and then constructs a geodesic distance matrix that provides a suitable and noise-free similarity for characterizing the intra-view proximity. Based on the geodesic distance matrix, the average similarity among all the views can be obtained, which is used in the manifold regularization term of the multi-view intact space learning for preserving the geodesic similarity. Extensive experiments have been conducted to evaluate the effectiveness of the proposed method, including the clustering experiments on 5 cancer datasets from the cancer genome atlas and the classification experiments on 1 EEG dataset.