Large-scale reverse engineering by the Lasso

Abstract

We perform a reverse engineering from the ``extended Spellman data'', consisting of 6178 mRNA levels measured by microarrays at 73 instances in four time series during the cell cycle of the yeast Saccharomyces cerevisae. By assuming a linear model of the genetic regulatory network, and imposing an extra constraint (the Lasso), we obtain a unique inference of coupling parameters. These parameters are transfered into an adjacent matrix, which is analyzed with respect to topological properties and biological relevance. We find a very broad distribution of outdegrees in the network, compatible with earlier findings for biological systems and totally incompatible with a random graph, and also indications of modules in the network. Finally, we show there is an excess of genes coding for transcription factors among the genes of highest outdegrees, a fact which indicates that our approach has biological relevance.