Title: "A Bayesian approach for deconvolution of sparsely convolved nucleosome signals in yeast"
Speaker: Ji-Ping Wang, Department of Statistics, Northwestern Univeristy, Evanston, IL
Place: HORT 117; March 29, 2011, Tuesday, 4:30pm

Abstract

The accuracy of nucleosome map obtained using MNase is substantially affected by MNase specificity, known as a fact that MNase prefers to cleave into a dinucleotide consisting of A or T. Using a newly developed chemical method combined with high throughput sequencing (Flaus & Richmond, 1996, 1999; Boggard and Widom 2011), the mapping accuracy can be dramatically improved up to a single base pair. Evidence shows that in many local regions, the nucleosome positioning may be not unique. The overlap between neighboring nucleosomes causes convolution of positioning signals from the chemical method. To recover the true nucleosome map from the observed data requires deconvolution of these nucleosome positioning signals, which however is challenged by two factors: (1) the number of parameters in the model is huge; (2) the distance between overlapped nucleosomes and the number of overlapped nucleosomes can be arbitrary. In this talk, we tackle this deconvolution problem by a Gibbs sampler approach. We first train a 8-position chemical cut model based on the experimental data. At each position, the observed chemical cut frequency is assumed to follow a Poisson distribution. We refine the model iteratively based on 2000 best nucleosomes selected genome-wide. We assume that each genomic position can be the candidate for a nucleosome center (dyad), the likelihood for which is measured by a parameter defined as the dyad occupancy. The dyad occupancy parameter is estimated by the posterior mean from a Gibbs sampler at convergence. This approach not only establishes a set of unique nucleosomes with best dyad occupancy with stunning features compared with MNase based map, but also provides a comprehensive atlas of all possible positions of nucleosomes (with possible overlap) in the yeast genome.

Associated Reading:
1. Segal et al. 2006. A genomic code for nucleosome positioning link. Nature 442:772-778.

2. Wang et al. 2008. Preferentially quantized linker DNA lengths in Saccharomyces cerevisiae. PLoS Comput Biol 4(9): e1000175. doi:10.1371/journal.pcbi.1000175

3. Xi et al. 2010. Predicting nucleosome positioning using a duration Hidden Markov Model. BMC Bioinformatics 2010, 11:346



Click here for a full schedule of BIOINFORMATICS SEMINARS, past and present.