Learning the transcriptional regulatory code

Mol Syst Biol. 5: 329Animal development is a fascinating process: starting from a single fertilized egg, an embryo grows and embryonic cells progressively differentiate into the diverse cell types and organs that make up an adult body. All this happens autonomously according to an intrinsic blueprint of development written in the four‐letter alphabet of the genomic DNA sequence.The genome not only encodes all developmentally important genes, but also carries the information necessary to specify the spatio‐temporal patterns of gene expression. The gene‐regulatory information is contained within the sequence of defined genomic regions, so‐called cis ‐regulatory modules (CRMs) or enhancers . These elements retain their cell‐type specific activity even when placed into an artificial context, for example when combined with a minimal promoter to drive expression of a reporter gene in transgenic animals (Arnone and Davidson, 1997).CRMs contain binding sites for specific sets of transcription factors (TFs) and are generally thought to integrate the bound factors’ regulatory cues, such that enhancer activity depends on the appropriate expression of the respective TFs. The simplicity of this model is attractive and it has indeed been shown that removing TFs or disrupting their binding sites by specific mutations impairs enhancer function (Arnone and Davidson, 1997). Despite its apparent simplicity, this model implies an underlying regulatory code that determines the exact requirements for enhancer function. A strong argument for the existence of this code would be the demonstration that enhancer activity can be predicted solely from the enhancers’ TF‐binding patterns. Ideally, enhancers with known activities could be used to learn rules that would be able to correctly predict the activity of novel enhancers.In a recent study, Eileen Furlong and colleagues follow …