Abstract:
|
Technologies have been developed to assess 3D interactions mediated by a specific protein, for example, ChIA-PET and Hi-Chip. A few methods that analyze ChIA-PET/Hi-Chip data to identify true 3D interactions have been developed. However, their relative performances are not so clear especially because there has not been a satisfying simulation protocol for in-silico studies. Also, recent methods are not utilizing adequate genomic annotations to facilitate the detection of true interactions. Here we propose a data integrated Bayesian mixture model to detect the true 3D interactions and a realistic simulation algorithm to evaluate the performance of methods. The simulation algorithm mimics the procedure of the ChIA-PET and Hi-Chip experiments step by step. The model allows information from 3D interactions’ upstream and downstream biological procedure, for example protein-protein interactions and neighboring gene expression level, to be integrated and improve the detection accuracy. Simulation study using the realistic simulation algorithm showed a better performance in power with well-controlled false positive rate by comparing to other methods.
|