Figure 1

Potential errors and biases of methylC-seq and binomial method. (A) Errors associated with the methylC-seq method. Non-methylated Cs may not be completely converted (non-conversion error, non-methylated C remains as C). In addition, methylated Cs may undergo conversion (over-conversion error, methylated C converts to T). (B) Reduced power of the binomial test in sparsely methylated genomes and low coverage. The Y-axis indicates FDR-corrected q-values from the binomial test, calculated following the equation (2) in the main text. The X-axis indicates the percentiles of p-values, which is equivalent to the whole genome methylation levels. Four cases are shown, including when a specific cytosine is covered by a single ‘C’ read (filled circles), one ‘C’ and one ‘T’ reads (crosses), one ‘C’ and two ‘T’ reads (open triangles) and two ‘C’ reads (filled triangles). The fractional methylation levels of these four cases are all substantial, 100%, 50%, 33% and 100%, respectively. However, in sparsely methylated genomes, many of these sites will have q-value > 0.05 and will be classified as ‘unmethylated’. For example, when only a single ‘C’ read is available (line with filled circles), despite the fact that the read itself indicates a 100% methylation, it will be designated as unmethylated (q-values > 0.05) unless the overall methylation level of the genome is greater than 4%. In another case, when a C is covered by one ‘C’ Read and two ‘T reads (line with open triangles), the fractional methylation level of such a position is 33%. However, such a site will be called as ‘unmethylated’ unless the overall level of methylation in the genome is 12% or higher.