Fig. 1

Overview of the diffBUM-HMM computational analysis pipeline. For each experimental condition (e.g. conditions 1 and 2), the log-ratios of drop-off/mutation rates (LDRs/LMRs) at each nucleotide position are computed for pairs of control samples to give a null distribution, in order to quantify variability in drop-off or mutation rates observed by chance. LDRs/LMRs are also computed similarly for all possible treatment-control comparisons. Coverage-dependent biases are then removed by applying a variance stabilization transformation. Subsequently, per-nucleotide empirical P values are computed for all possible treatment-control comparisons in each condition, by comparing the corresponding log-ratios to the null distribution. diffBUM-HMM is run on P values associated with the two independent conditions as observations, leaving out any nucleotides with missing data. The resulting output is a posterior probability of modification for each nucleotide, ranging from 0 to 1. diffBUM-HMM reports whether nucleotides were unmodified in both conditions, modified in either of the conditions or modified in both conditions