A computational method for predicting the location of a sequenced CpG island within its associated gene

CpG islands are short regions within mammalian DNA having a high G+C content and a nonsuppressed CpG dinucleotide frequency. Since all known CpG islands are located either within a gene or within close proximity to one, their presence is a landmark for a large number of genes within the genome. This aspect can be exploited by researchers characterizing and mapping the genome. Whereas most documented CpG islands are associated with the 5' regions of genes, a significant fraction are located downstream of the transcription initiation site. Whether the association of CpG islands with genes is fortuitous is not known; however, the information supplied by their presence in a sequence can be applied without this knowledge. The purpose of this work was to develop a computational means for distinguishing a CpG island located within the 5' region of a human gene from one located within the 3' body of a transcript based on its DNA sequence. This distinction would be valuable in extracting the maximum amount of information from either a genomic clone or an anonymous DNA sequence. The resulting CpG classifier was developed using a word preference based method and correctly distinguished 90.7% of a test set consisting of approximately one- third of all known human CpG islands. This tool should be useful for biologists for designing experiments and may provide insights into the function and derivation of the CpG islands.