Abstract

this paper, we propose a methodology for making sense of large, multiple time-series data sets arising in expression analysis, and evaluate it both theoretically and through a case study. First, we build a graph representing all putative activation/inhibition relationships by analyzing the expression profiles for all pairs of genes. Second, we prune this graph by solving a combinatorial optimization problem to identify a small set of interesting candidate regulatory elements. We do not assert that we identify "the" regulatory network as a result of this computation. However, we believe that our approach quickly enables biologists to identify and visualize interesting features from raw expression array data sets. We have implemented our methodology and applied it to the analysis of the Saccharomyces cerevisiae data set. In this paper, we report on our implementation and the results of our data analysis. The problem of inducing gene regulation networks has recently come to the computational biology community. Initial attempts at modeling gene expression abd programs to induce regulatory networks from data include [2, 10, 13]. To take fullest advantage of laboratory experiments that can be performed in which a given set of genes can be explicitly expressed or repressed, and the consequences of these genes on expression biologically determined, Akutsu, et.al. [1] considers the problem of designing a minimum-size series of experiments guaranteed to result in the identification of the correct regulatory network. The candidate regulatory network proposed by our system depends upon the specific optimization criteria employed in the second phase of our procedure, although our experiments suggest that the optimal network is surprisingly robust to changes in the objective function...

Citations