Kernel-Based Gene Regulatory Network Inference

Yi Shi*, Yuhong Guo, Guohui Lin, Dale Schuurmans

Department of Computing Science, University of Alberta, Edmonton, Alberta, T6G2E8, Canada. ys3@cs.ualberta.ca

Proc LSS Comput Syst Bioinform Conf. August, 2010. Vol. 9, p. 156-165. Full-Text PDF

*To whom correspondence should be addressed.

We propose a kernel-based method for inferring regulatory networks from gene expression data that exploits several important factors previously neglected in the literature, including expression clustering, nonlinear regulator-gene relationships, variable time lags and gene competition. In particular, our approach infers regulatory relationships by encouraging genes with similar expression patterns to share common regulators. Furthermore, since regulator-gene expression relationships are not typically linear but instead obey a wider class of canalyzing relationships, we map the relations between transcription factors to an implicit higher dimensional space that is able to model more complex interactions. Such a kernel-based approach avoids explicit enumeration of canalyzing regulations while allowing nonlinear relations between transcription factors to be discovered. Third, to address the problem of varying regulation time lags, we exploit cubic spline interpolation to identify more precise lag times from discretely sampled expression levels, leading to more robust inference of regulatory delays. Finally, we model competition between genes; an effect that has not been explicitly modeled by previous methods. The combination of these extensions leads to more accurate inference of regulator-gene causal relations from gene expression data.