Background: The liver X receptor (LXR), a transcription factor that forms a heterodimer with the retinoid X receptor, plays a key role in the transcriptional regulation of many important genes implicated in prevalent metabolic diseases. In spite of numerous studies, a complete list of LXR direct target genes remains elusive. To complement experimental approaches, computational prediction can be used to help build such a list because all LXR target genes are expected to carry the response elements (LXREs) in their promoter or enhancer regions. In practice, however, such a prediction has been hampered by the inaccuracies of currently available predictive models of LXREs. We report on a novel computational application for the highly accurate prediction of LXREs in DNA sequences.
Methods: We first conducted a comprehensive review of experimentally determined LXR target genes and collected all known LXREs. Subsequently, all such sites were classified using various computational methods based on sequence similarity to identify multiple subtypes. A library of Hidden Markov Models (LXRE.HMM) was developed to represent all subtypes and to enable the promoter scanning of LXR target genes.
Results and conclusion: Our model outperformed the widely used LXRE model in MatInspector in identifying the LXREs for all known LXR direct target genes at the experimentally verified positions. As a result, this new approach will make the genomewide prediction of LXR target genes feasible.