A group sparse joint nonnegative matrix factorization model for multiple data integration

In many disciplines, recent advances in high-throughput omics technologies have enabled researchers to collect large-scale datasets on a common set of samples. Integrating the data sets from different platforms has the potential to enhance knowledge discovery. Most of the existing methods focus on two data integration, such as canonical correlation analysis (CCA) based model and partial least squares (PLS) based model. Joint nonnegative matrix factorization simultaneously projects multiple data matrices onto a common subspace to explore the correlative relationship, but it ignores the prior knowledge or structure information within the datasets. In this paper, we incorporate the group sparse regularization into the JNMF model and propose a group sparse JNMF (GSJNMF) model. In the experiments, we design three data matrices with group information and some group components are correlated. We compare GSJNMF with JNMF on the simulated data and the performance verifies the effectiveness of the proposed method.