Systems biology A Stepwise Structural Equation Modeling Algorithm to Recon- struct Genetic Networks

Motivation: A stepwise structural equation modeling algorithm (SSEM) has been developed to infer genetic networks from time course microarray data. Besides observed variables, SSEM also incorporates hidden variables in order to depict regulations from proteins and other molecules that are not measured by microarrays. SSEM can learn the structure of a genetic network from microarray data. We have simulated data from a 6-gene and a 10-gene network, under median to high noise levels, to determine with which criterion SSEM works best out of the six goodness-of-fit indices studied. Next, we have applied SSEM to real microarray data in yeast (with no replicates) to infer transcriptional compensation interactions among six genes. Results: SSEM with BIC results in the highest true positive rates, the largest percentage of correctly predicted links from the total number of existing links, and the highest true negative (non-existing links) rates. We have applied SSEM with BIC to reconstruct a 6gene network in yeast, and compared the results with those obtained using three Bayesian network algorithms from Beal et al. (2005), Rangel et al. (2004) and Perrin et al. (2003). The modified true positive rates of SSEM and the former two algorithms are about 56%, 13% and 44%, respectively; while LDS of Perrin et al. (2003) can not confirm the existence of any links. Contact: [email protected] Availability: Supplementary data is available at http://www.stat.siniac.edu.tw/~gshieh/ssem.htm.