Statistical Models for Detecting Differential Chromatin Interactions Mediated by a Protein

• Published in: PloS one Vol. 9; no. 5; p. e97560
• Main Authors: Niu, Liang, Li, Guoliang, Lin, Shili
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 16.05.2014; Public Library of Science (PLoS)
• Subjects: Analysis; Androgens; Binding sites; Bioinformatics; Biology and Life Sciences; Chromatin; Chromatin - metabolism; Chromosomes; Computer Simulation; Crosslinking; Data processing; Deoxyribonucleic acid; Developmental stages; DNA; Experiments; Formaldehyde; Gene expression; Gene sequencing; Genomes; Genomics; Humans; Kinases; Mathematical models; Models, Chemical; Models, Statistical; Physical Sciences; Positron emission; Prostate cancer; Proteins; Proteins - metabolism; Software; Statistical analysis; Statistical models; Studies; Tomography; United States > US; Ohio
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0097560

Chromatin interactions mediated by a protein of interest are of great scientific interest. Recent studies show that protein-mediated chromatin interactions can have different intensities in different types of cells or in different developmental stages of a cell. Such differences can be associated with a disease or with the development of a cell. Thus, it is of great importance to detect protein-mediated chromatin interactions with different intensities in different cells. A recent molecular technique, Chromatin Interaction Analysis by Paired-End Tag Sequencing (ChIA-PET), which uses formaldehyde cross-linking and paired-end sequencing, is able to detect genome-wide chromatin interactions mediated by a protein of interest. Here we proposed two models (One-Step Model and Two-Step Model) for two sample ChIA-PET count data (one biological replicate in each sample) to identify differential chromatin interactions mediated by a protein of interest. Both models incorporate the data dependency and the extent to which a fragment pair is related to a pair of DNA loci of interest to make accurate identifications. The One-Step Model makes use of the data more efficiently but is more computationally intensive. An extensive simulation study showed that the models can detect those differentially interacted chromatins and there is a good agreement between each classification result and the truth. Application of the method to a two-sample ChIA-PET data set illustrates its utility. The two models are implemented as an R package MDM (available at http://www.stat.osu.edu/~statgen/SOFTWARE/MDM).