Template-based models for genome-wide analysis of next-generation sequencing data at base-pair resolution

We consider the problem of estimating the genome-wide distribution of nucleosome positions from paired-end sequencing data. We develop a modeling approach based on nonparametric templates to control for the variability along the sequence of read counts associated with nucleosomal DNA due to enzymatic digestion and other sample preparation steps, and we develop a calibrated Bayesian method to detect local concentrations of nucleosome positions. We also introduce a set of estimands that provide rich, interpretable summaries of nucleosome positioning. Inference is carried out via a distributed HMC algorithm that can scale linearly with the length of the genome being analyzed. We provide MPI-based Python implementations of the proposed methods, stand-alone and on Amazon EC2, which can provide inferences on an entire S. cerevisiae genome in less than 1 hour on EC2. We evaluate the accuracy and reproducibility of the inferences leveraging a factorially designed simulation study and experimental replicates. The template-based approach we develop here is also applicable to single-end sequencing data by using alternative sources of fragment length information, and to ordered and sequential data more generally. It provides a flexible and scalable alternative to mixture models, hidden Markov models, and Parzen window methods.