Skip to main content
. 2008 Apr 15;36(10):e55. doi: 10.1093/nar/gkn122

Table 1.

Functions for simulating experimental methods of DNA methylation mapping in silico

Method name References Method type Comment Simulation function
Differential methylation hybridization (DMH) Huang et al. (28) Khulan et al. (29) Pfister et al. (30) Methylation-specific fodigestion, qualitative Quantification is difficult due to different oligomer affinities and DNA melting temperatures

  • A1: HiMeth if #CpGpattern* & meth ≥ 50% ≥3

  • A2: HiMeth if #CpGpattern* & meth ≥ 50% ≥2

  • A3: HiMeth if #CpGpattern* & meth ≥ 50% ≥1 *pattern in {ACGT, CCGC, CCGG, GCGC}
Sequencing of methylation- specific digestion products Rollins et al. (31) Methylation-specific digestion, quantitative Quantification is possible if sequencing depth is high

  • B1: Profile(all CpGs in ACGT patterns)

  • B2: Profile(all CpGs in CCGC patterns)

  • B3: Profile(all CpGs in CCGG patterns)

  • B4: Profile(all CpGs in GCGC patterns)

  • B5: Profile(all CpGs in all four patterns)
Methyl-DNA immunoprecipitation plus tiling microarrays (MeDIP-chip) Weber et al. (32) Weber et al. (23) Zhang et al. (33) Zilberman et al. (34) Immunoprecipitation, qualitative Quantification is difficult due to different oligomer affinities and DNA melting temperatures

  • C1: HiMeth if #CpGmeth ≥ 67% ≥4*

  • C2: HiMeth if #CpGmeth ≥ 50% ≥3*

  • C3: HiMeth if #CpGmeth ≥ 33% ≥2* *minimum value per 200 bp
Sequencing of MeDIP-generated DNA libraries (MeDIP-seq) Established at several labs, e.g. at the Max Planck Institute for Molecular Genetics (H. Lehrach, personal communication) Immunoprecipitation, quantitative Quantification is possible if the enrichment is statistically corrected for local differences in CpG density

  • D1: Value(Mean(all CpGs))

  • D2: Value(Median(all CpGs))
Microarray hybridization of bisulfite-converted DNA Adorjan et al. (35) Gitan et al. (36) Kimura et al. (37) Yan et al. (38) Bisulfite conversion, qualitative Quantification has been attempted but is often unreliable

  • E1: HiMeth if mean (all CpGs) ≥ 67%

  • E2: HiMeth if mean (all CpGs) ≥ 50%

  • E3: HiMeth if mean (all CpGs) ≥ 33%
Direct sequencing of bisulfite-converted DNA Eckhardt et al. (15) Lewin et al. (19) Rakyan et al. (39) Bisulfite conversion, quantitative Quantitative and applicable to either all CpGs of an amplicon (by Sanger sequencing) or to a subset (by primer extension or pyrosequencing)

  • F1: Profile(all CpGs)

  • F2–F5: Profile(1 to 4 random CpGs)

  • F6: Profile(center CpG)]

  • F7: Profile(first and last CpG)

  • F8: Profile(CpGs at positions 1/3 and 2/3)

  • F9: Profile(first, center and last CpG)

  • F10–F20: Profile (0%, 10%, … , 100% of CpGs, rounded to the next integer value and randomly selected)
Rule-based guess (for comparison as a negative control) None No DNA methylation data is taken into account Worst-case baseline that any method should compare favorably with

  • G1: Value(0% methylated)

  • G2: Value(50% methylated)

  • G3: Value(100% methylated)

  • G4: Value(LowMeth)

  • G5: Value(MeanMeth)

  • G6: Value(HiMeth)

  • G7: Profile(random methylation values)

This table summarizes the experimental methods for DNA methylation mapping that are covered in this study, and it describes the functions that were constructed to simulate them in silico (rightmost column). The simulation functions are written in an abbreviated notation, as if-clauses, as profile statements or as value assignments. (i) For if-clause rules, a methylation constant named HiMeth is assigned to all CpGs in amplicons identified as high-methylation and a constant named LowMeth is assigned to all CpGs in low-methylation amplicons. We set HiMeth = 80.39% and LowMeth = 13.13%, which are the mean methylation levels of all amplicon that exceed or fall below 50% methylation, respectively, in the HEP dataset. (ii) For profile statements, a subset of CpGs that fulfill the condition in brackets are selected and the methylation values of all unselected CpGs are determined by interpolation or extrapolation. (iii) Value assignments are a special case of profile statements, in which no CpGs are selected and the methylation values of all CpGs are set to a constant value (MeanMeth = 56.91% for the HEP dataset). #CpGcondition stands for the number of CpGs in the amplicon that fulfill the condition. The source code implementing each of these rules is available on request (written in the Python programming language).