Figure 5
From: Normalization and experimental design for ChIP-chip data
Illustration of the key idea. When the log-ratio profile shows binding for a large fraction of probes (a), the distribution of log-ratios has a heavy tail on the right side (b). But the profile of the differences between neighboring probes (c) are not sensitive to the amount of binding. The few points that correspond to the jumps in (a) are large but are now distributed evenly between positive and negative values, and the distribution of the differences is symmetric (d).