Title: A sparse tensor subspace method for identifying biological modulators based on multilayer gene network analysis
Authors: Rui Yamaguchi - Tokyo University (Japan)
Seiya Imoto - University of Tokyo (Japan)
Satoru Miyano - Tokyo University (Japan)
Heewon Park - Hiroshima University (Japan) [presenting]
Abstract: Identifying crucial biological modulators has drawn a large amount of attention in precision medicine to understand molecular-cellular characteristic of disease. To identify biological modulators (e.g., candidate anti-cancer drug), we consider tensor similarity test for significant modulator-specific characteristic of disease. We consider drug sensitive and resistant gene network tensors, and then measure distance between the two tensors on tensor subspace. Although, the high dimensional genomic data include noisy features inevitably and the noise disturbs the process of not only similarity test but also subspace identification, relative little attention was paid to incorporating sparsity into tensor subspace method. In order to efficiently construct drug sensitivity-specific tensor subspace, we propose a novel sparse common component analysis based on L1-type regularization. By incorporating sparsity into subspace identification, our method constructs tensor subspace based only on the crucial common edges of multilayer gene networks without disturbance of noise. Thus, we can effectively extract characteristic of gene regulatory system in drug sensitive and resistant cell lines. We then propose a statistical test based on the similarity measure of tensors on the constructed tensor subspace to identify biological modules. The proposed method is applied to identify candidate anti-cancer drugs based on drug sensitive and resistance gene network tensors.