Unsupervised Learning of Finite Mixture Models
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
Sprout is a novel generative model for ChIA-PET data that characterizes physical chromatin interactions and points of contact at high spatial resolution. Sprout improves upon other methods by learning empirical distributions for pairs of reads that reflect ligation events between genomic locations that are bound by a protein of interest. Using these learned empirical distributions Sprout is able to accurately position interaction anchors, infer whether read pairs were created by self-ligation or inter-ligation, and accurately assign read pairs to anchors which allows for the identification of high confidence interactions. When Sprout is run on CTCF ChIA-PET data it identifies more interaction anchors that are supported by CTCF motif matches than other approaches with competitive positional accuracy. Sprout rejects interaction events that are not supported by pairs of reads that fit the empirical model for inter-ligation read pairs, producing a set of interactions that are more consistent across CTCF biological replicates than established methods.