Restriction landmark genomic scanning (RLGS) has been used to study aberrant CpG island methylation in cancer for more than ten years. This approach remains one of the most reliable ways to characterize CpG island hypermethylation in cancer and has been used both to characterize differences in aberrant methylation phenotypes and also to identify tumor suppressor genes. Not only have known tumor suppressor genes like Cdkn2a (p16), Itga4 (α 4-integrin) , and Igfbp7  been identified as targets of aberrant methylation in cancer by RLGS, but also novel tumor suppressor genes such as TCF21 , SLC5A8 , ID4 , BMP3B , and SOCS1  have been identified by RLGS.
RLGS is a two-dimensional gel electrophoresis method  that allows detection of DNA methylation if a methylation sensitive landmark enzyme such as NotI is used. Up to 2,000 end-labeled landmark sites are displayed in a single RLGS profile. The labeling of the sites is based on incorporation of radionucleotides into the NotI half-site by DNA polymerase. Methylated sites are not digested and are therefore not labeled, thus they do not contribute to the two-dimensional pattern of RLGS fragments. Spots present in a normal profile, but absent from a tumor profile are indicative of methylation of the landmark site. Furthermore, the profiles are quantitative such that spot intensity directly correlates with the degree of methylation of a locus, with partial methylation representing the cellular heterogeneity of the cancer both among the malignant cells and the associated non-malignant cells. RLGS profiles are highly reproducible allowing for comparison of different tissues (e.g. normal versus tumor tissue) or of different individuals. Although this approach does not cover every CpG island in the genome, as microarray-based approaches may achieve [9, 10], RLGS provides highly reliable data due to the molecular simplicity of the assay. Unlike microarray-based techniques that depend upon molecularly complex manipulations such as whole genome PCR, hybridization, and immunoprecipitation of methylated DNA, RLGS simply relies upon restriction digestion of DNA and gel electrophoresis.
Crucial to the successful implementation of RLGS for a CpG island methylation profiling study is the ability to identify the sequence of the targets of aberrant methylation, which has been a bottleneck in the flow of these types of studies. Significant improvements were made in RLGS spot cloning ability with the advent of arrayed boundary libraries, where restriction fragments from the arrayed library were mapped to the pattern of RLGS spots [11–13]. Although highly successful, this method is limited by the tremendous investment in effort required for each genome of interest and each enzyme combination of interest.
With the completion of the human and mouse genome sequences, a bioinformatics approach to RLGS fragment identification became possible where each restriction fragment's migration in both dimensions could be predicted to create a virtual RLGS profile. Such an approach could be applied to any sequenced genome, using any enzyme combination, and is unaffected by any potential tissue specific methylation. Furthermore, such an approach should be far less labor intensive to develop and use than the arrayed boundary library spot cloning method. The relative ease of setup and use, plus the incredible flexibility in the virtual approach, makes such a development highly attractive, allowing any researcher the ability to identify RLGS spots without the creation of specialized reagents.
This idea was partially realized in 2001 with a first generation version of a virtual RLGS gel that was called "Virtual Genome Scan" and applied to the human genome . In this system, the restriction fragment migration prediction formula was based solely on fragment length. Based on known fragment lengths and spot positions in both dimensions of rDNA and EBV genome derived spots, a cubic polynomial was derived giving fragment length as a function of spot location. By comparing the known positions of 22 previously identified spots with the positions of their corresponding fragments on the virtual gel, it was reported that 75% (16/22) of fragments were in a 12 × 30 pixel window (0.5 cm × 1.5 cm) around the predicted location, with the remaining fragments in a 24 × 128 pixel window (1.0 cm × 6.4 cm) . This approach was used to identify one previously unknown RLGS spot methylated in neuroblastoma and found to be in the 5' end of the ALX3 gene . Taking a similar approach, but incorporating the known chromosome of origin of the RLGS spots  into the predictions, Zardo et al (2002)  determined 96 previously unknown spot identities, and Hong et al (2003)  determined approximately 100 additional new spot identities.
A second virtual RLGS system (viRLGS) was reported in 2003 and applied to the arabidopsis and mouse genomes . The migration algorithms used were similarly based solely on the length of the DNA fragments. In the mouse virtual profile it was observed that there were 710 more spots on the virtual profile than are found on the same region of the actual profile. Given that in the mouse genome, unlike the human genome, many of the landmark sites (NotI sites) are within repetitive elements like LTRs, Repeat Masker was able to remove nearly 500 of these extra spots from the virtual profile . This approach was used to identify tissue specifically methylated RLGS spots in mouse  and to identify CpG islands that become hypomethylated in mouse ES cells when DNA methyltransferases are knocked out .
With approximately 2500 spots on a 35 × 43 cm gel in the human NotI-EcoRV-HinfI enzyme combination most commonly used, the average distance between neighboring spots is 0.7 cm. To generate a virtual image that approximates the true image, such that the relative predicted positions of any two neighboring spots does not flip-flop, the window of error for spot prediction should be, at most, half this average distance or 0.35 cm. One confounding issue is that others have demonstrated a considerable correlation between DNA fragment curvature and gel mobility [22–25]. Thus, it is clear that, while a simple fragment migration model based upon fragment length alone is a good start, it must be improved upon. In addition, since in the human genome NotI sites rarely cut in repetitive elements, use of a repeat masker does not help to remove the extra spots in a human virtual RLGS profile.
We have developed a novel system of virtual RLGS (vRLGS) using first and second dimension migration algorithms that use the sequence characteristics of each fragment along with the fragment length to determine spot position. These algorithms were determined based on the identification of 795 RLGS spots in the human NotI-EcoRV-HinfI profile. In addition, we have defined a filter to remove extra spots from both the human and mouse vRLGS profiles based on GC content. We report greatly improved accuracy of pattern prediction even when the migration algorithms are applied to different enzyme combinations and different genomes from which they were derived. These improvements are combined with a novel graphic interface that allows for visualization of virtual spots that 1) look like real spots, and 2) allow the user to overlap the virtual and real profiles. These changes make the bioinformatics approach to identifying RLGS spots of interest a highly viable and effective alternative to more complex RLGS spot cloning methods.