A physical model for tiling array analysis

Abstract

Motivation: Chromatin immunoprecipitation (ChIP) is a powerful experimental approach to identify in vivo binding sites of sequence-specific transcription factors (TFs). These experiments are designed to specifically enrich DNA fragments that are bound to the TF. Tiling arrays have become more and more popular for the identification of these DNA fragments. However, many studies showed that only a fraction of the identified DNA fragments contains bona fide binding sites for the TF, suggesting that indirect binding mechanisms play a very important role. We explored the possibility that the lack of binding sites can also be explained by problems in identifying ChIP-enriched DNA fragments from the measured intensities. Results: We derived a physical model that explains some (but not all) variation of the measured probe intensities of Affymetrix tilling arrays. We used the physical model to estimate the probe-specific behavior and corrected for it. Subsequently, we developed a method to identify ChIP-enriched DNA fragments. We termed it physical model for tiling array analysis (PMT). We applied PMT to the data of ChIP-chip experiments interrogating chromosome 21 and 22 of the human genome for binding of the TFs MYC, SP1 and P53. Almost all regions recovered by PMT showed evidence for sequence-specific binding of the TFs. © 2007 The Author(s).