The framework of our method is derived from a physical comprehension of the ChIP-seq process. (A) Illustration of chromatin immunoprecipitation steps. A DNA representation (blue string) contains multiple binding sites (light purple) that may or may not be bound by their corresponding transcription factor (purple hexagon). The pink dashed boxes highlight a DNA fragment that is simultaneously bound by two TFs. This fact motivates the double-binding signal. The black dashed box highlights the break point selection caused by immunoprecipitation. Many break points (solid dark circles) occur, but only the one at each edge that is closest to the binding site is selected for sequencing. This indicates that the impulse response follows an extreme value distribution (see main text). (B) Detailed representation of the impulse response. At each strand ([green] positive stand, [pink] negative strand), the impulse response follows a distribution f(x) that represents the distance from the DNA fragment edges to the center of the binding sites. The coverage at each strand is separated by a peak shift that is equal to twice the distance of the binding site center and the maximum of f(x). The strand-specific coverage and peak shift is a consequence of the directionality of sequencing. (C) Representation of the double-binding signal. At each strand, the impulse response follows the same distribution f(x) observed for the single-binding signal. On the positive strand, it refers to the binding site closer to the 5′ end; on the negative strand, it refers to the binding site closer to the 3′ end. This causes an additional peak shift equal to the distance between the binding sites. (D) The effect of the double-binding signal in the signaling process. A region with two binding sites contains three sources of signal: a single-binding signal (solid blue upward arrow) for each binding site and also a double-binding signal (dashed blue upward arrow, left). The decomposition of the coverage into individual impulse responses and the corresponding impulse representation are also shown (right).
