Model based identification of transcription factor activity from microarray data

With the increase in volume of gene expression data available from high throughput microarray experiments, much research interest has been directed at building mathematical models of the process of gene regulation. Such models have primarily been used for the so called reverse engineering of regulatory networks; inferring possible regulatory interactions directly from microarray data, for example [1–4]. By using microarray data, all of these techniques make the implicit assumption that there is a direct relationship between the level of mRNA of genes coding for transcription factors (TFs) and the mRNA levels of their gene-targets. Whilst for some TF-gene pairs, this is likely to be a reasonable assumption, there are many examples of regulatory interactions where it is not due to modifications of the TF after translation. Such modifications cannot be measured on the microarray leading to minimal correlation between the expression levels of the TF gene and it’s targets. It is obvious therefore that any models of regulation encoding a direct relationship between the mRNA levels of the two genes will be highly inaccurate over a wide range of interactions and conditions.