Sparse Gaussian graphical model with missing values

Recent advances in measurement technology have enabled us to measure various omic layers, such as genome, transcriptome, proteome, and metabolome layers. The demand for data analysis to determine the network structure of the interaction between molecular species is increasing. The Gaussian graphical model is one method of estimating the network structure. However, biological omics data sets tend to include missing values, which is conventionally handled by preprocessing. We propose a novel method by which to estimate the network structure together with missing values by combining a sparse graphical model and matrix factorization. The proposed method was validated by artificial data sets and was applied to a signal transduction data set as a test run.