Human genes are selectively expressed in cells dependent upon the local conditions defined by physiological, developmental and environmental signals. Under the appropriate conditions, genes are transcribed to produce RNA - initiating the flow of information from static DNA sequences into active proteins. Regulation of transcription is mediated by the interaction of sequence-specific DNA binding proteins (transcription factors) with target sites within genes. Recently it has become apparent that human genes contain regulatory modules - locally dense clusters of binding sites for subsets of transcription factors. This has opened new opportunities in computational biology to expand our understanding of the circuitry of human gene regulation.

Transcription factors tolerate considerable variation in their target sequences, which can be captured in probabilistic models. These models are effective at predicting test tube interactions, but most predicted binding sites have no function within living cells. Based on the module model, we have developed methods to detect clusters of binding sites that are predictive for function. Key problems in this area include the assessment of the significance of pattern clusters in DNA sequence and the acceleration of procedures to facilitate efficient screens of the available genome sequences.

Reliable reference data collections are sparse in biology. Only a few contexts are studied with sufficient depth to build predictive models. The greatest challenges are found in extracting principles from the sparse examples that enable the de novo discovery of new information about partially characterized systems. Using a Gibbs sampling procedure coupled to comparative analysis of the human and mouse genome sequences, it is possible to extract the properties of regulatory modules for under-studied biological contexts.

Reference: Lenhard B, Hayes WS, Wasserman WW. (2001). GeneLynx: a gene-centric portal to the human genome. Genome Res. Dec; 11(12):2151-7.

The paper is available at http://www.genome.org/cgi/content/full/11/12/2151