Largescale reverse engineering by the lasso

What we do Utilize the Lasso method (Tibshirani 1996) to obtain effective regu-lator networks for genes in yeast from time series microarray data, both for the full set of 6178 genes and the 800 cell-cycle regulating genes (Spellman et al. 1998). Analyze some topological properties of the obtained network. Study how local properties of nodes in the network and biological properties of the genes correspond. What we conclude The network analysis reveal a broad distribution of outdegrees and a moderate clustering. There seems to be a negative correlation of outdegree ranks to inviability of genes, and a positive correlation to nucleus localiza-tion. What is next? Include secondary effects, by in silico over expression experi-ments. Search for frequent regulatory motifs. The numerical method We estimate the missing values by the weighted KNN procedure (Troyanskaya et al. 2001). We assume a linear, continuous, deterministic model of the loga-rithm of the expression levels, _xi(t) = PN j=1 wijxj(t). The network is formed by minimizing the residual sum of squares, with the addition of the Lasso penalty term, i.e., for i = 1; : : :;N, min KX k=