A STATE SPACE REPRESENTATION OF VAR MODELS WITH SPARSE LEARNING FOR DYNAMIC GENE NETWORKS

• DOI: 10.1142/9781848165786_0006
• Publication Date: 2010-03-18T10:30:52Z
• Authors: KANAME KOJIMA; RUI YAMAGUCHI; SEIYA IMOTO; MAI YAMAUCHI; MASAO NAGASAKI; RYO YOSHIDA; TEPPEI SHIMAMURA; KAZUKO UENO; TOMOYUKI HIGUCHI; NORIKO GOTOH; SATORU MIYANO
• Keywords: State-space representation; Representation; Regularization

We propose a state space representation of vector autoregressive model and its sparse learning based on L1 regularization to achieve efficient estimation dynamic gene networks time course microarray data. The proposed method can overcome drawbacks the model; assumption equal interval lack separation ability observation systems noises in former modularity network structure latter method. However, simple implementation requires calculation large inverse matrices number times during parameter process EM algorithm. This limits applicability relatively small set. thus introduce new technique for algorithm that does not require matrices. is applied data lung cells treated by stimulating EGF receptors dosing an anticancer drug, Gefitinib. By comparing estimated with control using non-treated cells, perturbed genes drug could be found, whose up- down-stream may related side effects drug.