|
Abstract:
|
Advances in technology have led to generation of omics data (single transcriptomics data or multiple types of data such as genetics and transcriptomics data together), which in some studies may be collected from multiple conditions (e.g., cell types, tissue samples, disease states). Such data can be analyzed to explore co-expression network or expression quantitative trait loci (eQTL) or metabolomic quantitative trait loci (mQTL) networks across multiple conditions. However, integrative analysis of omics data across multiple conditions presents a number of challenges, analytical and numerical. To tackle these challenges, we propose a novel, computationally efficient approach based on multidimensional array (tensor) decomposition and regularization techniques. The proposed approach enables us to identify common network structures across multiple conditions as well as condition-specific network structures, while encouraging sparsity in latent factors used in the tensor decomposition. Synthetic data and real data are used to demonstrate the performance of our approach.
|
